Intrinsic dreams are not produced without REM sleep mechanisms: evidence through elicitation of sleep onset REM periods


Tomoka Takeuchi Ph.D. Department of Psychology, Brock University, St Catharines, Ontario, L2S 3A1, Canada. Tel.: + 1 905 688 5550, ext.4676; fax: + 1 905 688 6922; e-mail:


The hypothesis that there is a strict relationship between dreams and a specific rapid eye movement (REM) sleep mechanism is controversial. Many researchers have recently denied this relationship, yet none of their studies have simultaneously controlled both sleep length and depth prior to non-REM (NREM) and REM sleep awakenings, due to the natural rigid order of the NREM–REM sleep cycle. The failure to control sleep length and depth prior to arousal has confounded interpretations of the REM-dreams relationship. We have hypothesised that different physiological mechanisms underlie dreaming during REM and NREM sleep, based on recent findings concerning the specificity of REM sleep for cognitive function. Using the Sleep Interruption Technique, we elicited sleep onset REM periods (SOREMP) from 13 normal subjects to collect SOREMP and sleep onset NREM (NREMP) dreams without the confounds described above. Regression analyses showed that SOREMP dream occurrences were significantly related to the amount of REM sleep, while NREMP dream occurrences were related to arousals from NREM sleep. Dream properties evaluated using the Dream Property Scale showed qualitative differences between SOREMP and NREMP dream reports. These results support our hypothesis and we have concluded that although ‘dreaming’ may occur during both REM and NREM periods as previous researchers have suggested, the dreams obtained from these distinct periods differ significantly in their quantitative and qualitative aspects and are likely to be produced by different mechanisms.


Since the discovery of a relationship between rapid eye movement (REM) sleep and dreaming (Dement and Kleitman 1957), many researchers have investigated dream production mechanisms. Some form of mentation is likely to be reported by people awakened during various sleep states, such as sleep onset, NREM sleep, and REM sleep. However, whether there are different qualitative and quantitative aspects to these mentations during REM or NREM sleep has been debated (Nielsen 1999, 2000).

Physiological phenomena observed during REM sleep (ocular saccades and muscle atonia) are known to be regulated by a population of neurons whose activation is specific to REM sleep (Hobson et al. 1998, 2000). Gamma range activity during REM sleep (see the review of studies by Gottesmann 1999, p. 473–477) has also been observed. Decreased activity in dorsolateral prefrontal cortex specific to REM sleep has been reported in studies using functional MRI (Braun et al. 1998; Maquet et al. 1996). Functionally, it is during REM sleep that dreams are more likely to occur, visual imagery is more vivid, and some type of memory consolidation is likely to happen (De Koninck et al. 1989; Hobson et al. 2000; Karni et al. 1994; Plihal and Born 1997; Smith 1995; Stickgold et al. 1999). Considering these findings, it follows that dreams produced during REM sleep might be affected by the physiological mechanisms contributing to REM sleep.

On the other hand, many researchers have claimed that dreaming occurs during both REM and NREM sleep and have rejected the strict association between REM sleep mechanisms and dreams (Antrobus 1983; Antrobus et al. 1995; Cavallero et al. 1992; Cavallero et al. 1990; Foulkes 1993; Foulkes and Schmidt 1983; Rosenlicht et al. 1994).

Nielsen (1999, 2000) has reviewed these contradictory findings concerning dream generation systems and has assigned them to one of two different categories: (1) the 1-generator model, which suggests that a single set of imagery processes produces sleep mentation regardless of the sleep stage in which it occurs; and (2) the 2-generator model in which REM and NREM sleep mentation reports originate from qualitatively different imagery generation systems. Furthermore, he proposed the ‘Covert (phantom) REM model’ to reconcile the two theories. In his model, sleep mentation, which is basically coupled with REM sleep processes, may occur during sleep onset or NREM sleep in a ‘covert’ manner under certain conditions depending on the temporal proximity of those states to REM sleep. It is suggested that the covert REM sleep process could produce dreamlike activities in various conscious states.

Until now, dream studies were limited by the natural order of the NREM-REM sleep cycle. Because REM periods typically appear cyclically after approximately 90 min of NREM sleep, it is not possible to examine REM periods without the possibly contaminating effects of the previous NREM period. Similarly, accumulated sleep time or sleep stage changes are confounded with NREM–REM ordering of stages. Thus, previous dream studies have not simultaneously controlled prior sleep length and depth of NREM and REM sleep.

The NREM–REM sleep cycle is stable and reliable except for some clinical abnormal appearances of sleep onset REM period (SOREMP). SOREMP is one of the main features of narcolepsy (Hishikawa and Kaneko 1965). It is also known that SOREMPs sometimes appear in normal sleepers when circadian or ultradian rhythms are disrupted, such as under a free-running schedule (Czeisler et al. 1980a, b; Zulley 1980), acute reversal of the sleep–wake cycle (Weitzman et al. 1970), a multiphasic sleep–wake schedule (Carskadon and Dement 1975, 1980; Lavie 1986), or after experimental sleep interruptions (Fukuda et al. 1987; Schulz and Tetzlaff 1982).

Drawing from these findings, Miyasita et al. (1989a) developed a technique to elicit SOREMP experimentally from normal sleepers (the Sleep Interruption Technique). A positive linear relationship between NREM duration prior to the sleep interruption and the probability of eliciting SOREMP has also been found (Miyasita et al. 1989a). Sasaki et al. (2000) found that the probability of SOREMPs was influenced by the circadian rhythm. Utilizing these systematic and predictable features of SOREMP, the Sleep Interruption Technique allows us to elicit SOREMPs for experimental investigation. Furthermore, this technique has been adapted as a method of simulating some of the symptoms of narcolepsy (SOREMP, sleep paralysis, and hypnagogic hallucinations) in normal individuals (Takeuchi et al. 1992, 1994).

A wide variety of mentation reports are likely to be obtained during sleep onset periods, from simple experiences such as visual imagery, auditory experiences and bodily sensations to complex and hallucinatory experiences (Nielsen 2000; Vogel 1991; Hayashi et al. 1999). Because sleep onset itself is not a monotonous process, but a dramatic, dynamic, and systematic one, its definition seems to differ depending on the measurements in which researchers are interested (Ogilvie 2000). Bosinelli 1991 pointed out that one possible explanation for conflicted findings in sleep onset mentation might be confounded definitions of mentation and interpretations of sleep onset among researchers. Analyses using more subtle criteria have shown a diversity of relationships between EEG variables and mentation during the sleep onset process (Hori et al. 1994; Hayashi et al. 1999). At the same time, mentation reported during sleep onset may be more reliable than reports collected after hours of sleep due to the retrospective nature of the data. During the sleep onset period, the point when the mentation report is obtained is close to the point when that mentation was actually produced. Hence, the physiological data obtained from an identified epoch during which dreaming actually occurred, can be more accurately located in a shorter time window during sleep onset.

Considering these findings, the Sleep Interruption Technique appears to give useful information to dream studies, because it allows us to examine mentation collected from both REM and NREM sleep at sleep onset without the confounds of prior sleep length and depth.

In this study, controlling pre-sleep processes using the Sleep Interruption Technique, we examined: (1) whether there are quantitative differences in mentation between SOREMP and sleep onset NREM periods (NREMP); (2) whether the same physiological markers can predict ‘dream’ recall in SOREMP and NREMP; and (3) whether there are qualitative differences between ‘dreams’ in either type of sleep.



Thirteen female undergraduates, aged 18–20 years, took part in our study. All were free of cataplexy, sleep attacks and psychosis. Informed consent was obtained after explanation of the procedures, including sleep interruption. Participants spent 7 consecutive nights in our lab with polysomnographic recording (first and second nights were adaptation nights; third night was baseline night; fourth to seventh nights were experimental nights). The Sleep Interruption Technique (Miyasita et al. 1989a; Sasaki et al. 2000) was employed during each experimental night. Each participant’s sleep was interrupted for one hour, after 40 min of NREM sleep had elapsed since the termination of the first (Early condition) or the third (Late condition) REM period in the NREM–REM sleep cycle.

Eight participants (Protocol 1, See Fig. 1) were awakened from the Early (2 nights) or Late (2 nights) condition. Ordering was counterbalanced among participants. The remaining five participants (Protocol 2) were always awakened from the Late condition during all four experimental nights. Based on the finding of Miyasita et al. (1989b), it was assumed that awakening participants after 40 min of NREM sleep in the fourth NREM–REM sleep cycle would elicit an approximately equal number of SOREMP and NREMP episodes.

Figure 1.

 Experimental design. Upper; adaptation and baseline nights. Lower; experimental nights.

During each interruption, participants completed a 40-min auditory vigilance task to maintain minimal arousal. Following this vigilance task, participants were allowed to return to sleep and were awakened again after 5 min had elapsed from either the first appearance of rapid eye movements (SOREMP) or a sleep spindle/K-complex indicative of stage 2 sleep (NREMP). In the event that the first rapid eye movements were observed after a spindle/K-complex, the participant was awakened 5 min after these rapid eye movements and this was considered a SOREMP episode. After being awakened, participants were asked about their subjective state prior to being awakened (see Dream data collection, below). This procedure was repeated two more times (Fig. 1). After these procedures, participants were allowed to sleep until their net sleep time reached 7.5 h.

Dream data collection

In each interruption, participants were awakened by a tone burst (45 db; duration, 100 msec; ISI, 200 msec), presented through earphones. They were then asked to fill out simple questionnaires regarding their mental states just prior to the tone stimulus. If they had any mental activities, they were required to rate their mentations using the Dream Property (DP) scale (Takeuchi et al. 1996). After completing the DP scale, they were asked to verbally report their mentations, which were recorded for later transcription.

Dream property scale

Takeuchi et al. (1996) constructed the DP scales to provide information regarding the basic property of dreams without the confound of narrative length. It assesses four factors based on 15 polar adjectives with seven-point Likert scales (standardized):

1 Bizarreness (realistic–unrealistic, usual–unusual, believable–unbelievable, and ordinary–extraordinary);

2 Evaluation (dark–bright, gloomy–cheerful, uncomfortable–comfortable, and disgusting–likable);

3 Impression (unclear–clear, ambiguous–explicit, forgettable–memorable, and foggy–vivid);

4 Activity (quiet–noisy, static–dynamic, and still–bustling).

These were extracted from 200 polar adjectives by factor analysis and were normalized using the data from 318 dreams collected in the sleep laboratory. Also, the validity of the DP scale was confirmed by examining the relationship between the DP scores and REM sleep physiological activities, such as eye movements and muscle activities. Furthermore, the DP scale was translated and developed into an English version of the DP scale and was shown to reflect EEG activities underlying dreaming process (Takeuchiet al., in press). There were seven versions of the DP scale, each contained the same polar-adjectives, but the order was randomized. These DP scales were counterbalanced inter- and intra-subject and placed on the bedside table before each night.

Definition of dreaming

Nielsen (1999) examined dream studies from 29 REM and 33 NREM studies from the past few decades and reported that the average REM recall rate was 81.8% (SD=8.7) while NREM recall rate was 42.5% (SD=21.0). Thus, dream recall rate seems to vary, particularly in NREM sleep, depending on the definition of dreaming. Because of potentially systematic variances caused by experimenter judgement (Herman et al. 1978), we left the definition of the mentation to each participants’ own criteria, expecting that the error variance caused by participants’ judgement would be more randomly distributed than the experimenter’s judgement.

Mentation reports were classified depending on whether participants themselves reported experiencing Dreams, Thoughts, Nothing (an absence of mentation), or Recall failure (having mental activity but not being able to recall the content). To make the contrast among independent variables clearer, only data with Dreams and Nothing were used in further analyses (See RESULTS).


Participants were monitored polygraphically with an electroencephalogram (EEG; Cz, Oz) referred to averaged mastoids (A1, A2), horizontal and vertical electrooculogram (EOG), and mentalis electromyogram (EMG), during both sleeping and interruption periods.

Sleep measurements

Each sleep stage was scored in 30 s epochs using standard criteria (Rechtschaffen and Kales 1968). Sleep onset was defined as the first 30 s epoch of continuous stage 1 or 2 sleep. For the purpose of this study, an episode (SOREMP/NREMP) was defined as the time from the sleep onset following the last sleep interruption to the last complete 30 s epoch before the next awakening.

For further analyses, the total duration of time spent in each sleep stage and the total duration and frequency of body movements were calculated in each SOREMP or NREMP episode. The frequency of stage shifts was also counted. Statistical analyses were performed using SPSS 9.0 (SPSS Inc., Chicago, USA).


Ratio of SOREMP and NREMP episodes

As indicated in Table 1, one awakening in the Late condition was eliminated because the participant woke up before the appearance of REMs, spindle wave, or K-complex wave.

Table 1.   Mentations obtained Thumbnail image of

In the Early condition, there were only 4 SOREMPs, while there were 44 NREMPs. Only data from the Late condition was used in the further analyses due to the small number of SOREMPs in the Early condition and possible confounds caused by the different circadian effects between Early and Late conditions (Sasaki et al. 2000).

Out of a total of 107 interruption awakenings from the Late condition, 42 SOREMP episodes (39.3%) and 65 NREMP episodes (60.7%) were observed. A total of 52 mentations were collected, including 40 Dreams, one Thought, one report of Sleep Paralysis, and 10 Recall failures. As seen in Fig. 2, 32 out of 42 (76.4%) SOREMP episodes and 8 out of 65 (12.3%) NREMP episodes contained Dreams. Thus, the presence of Dreams was indeed closely related to SOREMP, while Recall failure and Nothing reports were related to NREMP.

Figure 2.

 Classification of mental states obtained from SOREMP and NREMP.

Sleep measurements during SOREMP and NREMP episodes with/without dreams

Given that Dreams were reported from both NREMP and SOREMP episodes, we investigated whether these were produced by the same or different mechanisms as demonstrated by physiological markers during both types of episodes. To determine which physiological markers would predict Dream occurrence (Dream vs. Nothing), we used hierarchical multiple regression analyses based on the sleep measurements during SOREMP and NREMP episodes (Table 2).

Table 2.   Summary of sleep measurements during NREMP/SOREMP episodes with/without dreaming Thumbnail image of

Other types of mentations such as Thoughts, Sleep paralysis, and Recall failures were excluded due to their ambiguous nature. Dream recall (Dream or Nothing) was regressed on time spent in each sleep stage (REM, W, 1, 2 for SOREMP and W, 1, 2 for NREMP) as well as body movements for both SOREMP and NREMP episodes. To compensate for the differing sleep lengths, total time from sleep onset to awakening was partialed out on the first step. This accounted for 13% of the variance (R=0.36, P=0.007) in the NREM episodes and virtually none (R=0.05, P=0.75) in the SOREMP episodes. In SOREMP episodes, time in stage REM, stage W (awakening), stage 1, stage 2, and body movements were entered on the second step of each regression. In NREM episodes, time in stage W, stage 1, stage 2 and body movements were entered in the second step.

It is clear that the amount of stage W (awakening period) significantly (< 0.001) contributed to the likelihood of Dream recall in NREMP episodes, while the amount of stage REM significantly (P=0.008) contributed to prediction of Dream recall in SOREMP episodes (Table 3). That is, during sleep onset periods elicited by our protocol, Dream occurrence was strongly related to stage W in NREMP episodes, and to REM sleep in SOREMP episodes.

Table 3.   Semipartial correlation obtained by multiple regression analysis using the time spent in each sleep stage and body movements to predict dream recall Thumbnail image of

To be certain of the effect obtained in the regression analysis, the data were also analyzed, eliminating episodes contaminated by the intrusion of arousal factors. We removed SOREMP and NREMP episodes that contained arousal related factors (e.g. body movements, stage W, non-descending stage shifts such as 2, 2, 1, 2 or 1, 2, 1, W). Out of 107 SOREMP and NREMP episodes, 78 contained some type of arousal-related factors. The remaining 29 episodes without any arousal factors were composed of only three SOREMP episodes (7%) and 26 NREMP episodes (40%). Interestingly, all 3 SOREMP episodes were accompanied by Dreams (100%) while there were no longer any Dreams reported from 26 NREMP episodes (0%). A binomial test provides some indication of how unlikely this occurrence is. If we use the likelihood of 31 Nothing reports (none of Dream, Thought, and Recall failure) in 39 NREMP episodes with an arousal as our null hypothesis, the probability of there being 26 Nothing reports in 26 NREMP episodes is only 0.0026. This result is consistent with the findings we obtained in the regression analysis. Both suggest that different physiological markers are related to Dream appearance between SOREMP and NREMP.

Dream properties of SOREMP and NREMP episodes

If, as hypothesized, there are different mechanisms producing NREMP and SOREMP Dreams, then we could also expect qualitative differences between these two types of Dreams. To test this hypothesis, DP scores from 8 Dreams from NREM episodes and 32 Dreams from SOREMP episodes were compared. Due to the sample size and a lack of homogeneity of variances, Mann–Whitney U-tests were conducted using median DP scores.

As shown in Fig. 3, NREMP and SOREMP Dreams differed significantly on the Evaluation (U=67.00, z′=−2.078, < 0.05), the Impression (U=41.50, z′=−2.954, < 0.005) and the Activity (U=27.50, z′=−3.423, < 0.0001) scales. SOREMP Dreams were characterized as clearer, more explicit, more memorable and more vivid (Impression) as well as noisier, more dynamic, and more bustling (Activity) compared with NREMP Dreams.

Figure 3.

 Comparison of dream properties between SOREMP and NREMP. *< 0.05; ***< 0.005.


Validity of Sleep Interruption Technique

By manipulating the natural order of the NREM–REM sleep cycle using the Sleep Interruption Technique, we succeeded in collecting mentations from SOREMP and NREMP with their pre-sleep processes controlled. With this technique, we were able to make a thorough comparison of SOREMP and NREMP mentations without the confounding effects of previous NREM/REM sleep, accumulated sleep time, and NREM-REM stage ordering.

Possible influence on the results caused by the specificity of our protocol

It may be suspected that mentations reported in our study are specific to our protocol and are therefore atypical because they were obtained using the artificial REM and NREM periods at sleep onset elicited by manipulating the NREM–REM cycle. The specific characteristics of SOREMPs in normal individuals have been reported to be:

1 likely to occur on non-24-h sleep/wake schedules (Czeisler et al. 1980a, b; Weitzman et al. 1970; Carskadon and Dement 1980; Fukuda et al. 1987);

2 related to the enhanced REM sleep pressure or REM propensity (Sasaki et al. 2000);

3 likely to occur at the trough of the body temperature rhythm (Czeisler et al. 1980a, b; Carskadon and Dement 1980);

4 a necessary but insufficient condition for the sleep paralysis and/or hypnagogic hallucination to occur (Takeuchi et al. 1992, 1994).

These are phenotypic aspects of the REM sleep observed during SOREMP episodes. That is, SOREMP still has all the usual REM components such as rapid eye movements, muscle atonia, and desynchronized EEG. These physiological activities are known to reflect specific patterns of neural activities during REM sleep (Hobson et al. 2000). Hypothesizing dreaming as one of the REM functions, none of the four aspects above are necessarily opposed to our proposition that dreaming is one of the functions of sleep onset REM sleep. Therefore, our protocol can still be useful in examining dreaming as one of the functional aspects of REM sleep. Nonetheless, the results in our study should be carefully interpreted until more normative studies of the SOREMP process have been documented.

Likewise, one might suspect the specific nature of NREMP and SOREMP episodes during sleep onset periods in this study. One clear difference between the sleep onset periods after interruption in our study and those in the initial sleep onset is the REM potential (Sasaki et al. 2000). In normal individuals, SOREMP does NOT appear during initial sleep onset unless the preceding sleep-wake cycle is disrupted, whereas sleep onset after an interruption produced SOREMPs in this study. Hence, sleep onset periods in our study may be closer to the states relating to the ‘Covert REM process’ (Nielsen 2000). Considering the definition of SOREMP (REM latency of less than 25 min), NREMP episodes obtained in our protocol might have shown REM sleep if participants’ sleep were maintained without the forced awakening after 5 min of stage 2. This may be a possible explanation of the NREMP Dreams obtained in our investigation.

Quantitative aspects of dreaming

In this study, Dream recall rates were 76.4% in SOREMP and 12.3% in NREMP. We left the definition of the mentation open to participants’ judgement, expecting that the error variance caused by individual differences would not be as biased as that caused by the experimenters’ judgement. When every mentation such as Thought, Sleep paralysis, and Recall failure are included, mentation recall rates were 81.0% from SOREMP and 27.7% from NREMP. Nielsen (2000) summarized REM and NREM mentation recall rates using 35 studies since 1953. According to his review, the mean REM recall rates increased from pre1962 (76.0%(S.D.=11.5)) to post1962 (84.1% (S.D. =6.7)) as have mean NREM recall rates (18.4% (S.D.=15.4) for pre1962 to 50.9%(S.D.=15.5) for post1962). Drawing from numerous findings, Nielsen (2000) states, ‘Thus, at most 25%, but possibly as little as 12%, of NREM awakenings in susceptible subjects will produce reports of dreaming. The more elaborate forms of dreaming (‘apex’ dreaming – the most vivid, intense, complex forms of dreaming: e.g. nightmare, sexual, archetypal, transcendental, titanic, existential, lucid dreaming) are even less prevalent’. Our result was close to Nielsen’s pre 1962 rate, when only Dream was included. When all mentation was included, it was close to the post 1962 rate. As a whole, the Dream recall rates obtained in our study may imply either that the average notion of dreaming may be closer to the definition of ‘apex’ dreaming, or that dream recall rates are different between REM and NREM sleep when their preceding sleep process are controlled (or both). The results on mentation recall rate obtained in our study seemed to be within the range of findings of most dream research. Nevertheless the differential recall rates for SOREMP and NREMP awakenings remain clear.

Relationship between quantitative aspects of dreams and sleep stage

Regression analyses showed a clear relationship between the increased amount of stage W and Dream recall in NREMP episodes and the increased amount of stage REM and Dream recall in SOREMP episodes (Table 3). Furthermore, when only ‘pure’ SOREMP or NREMP episodes without any arousal intrusion were examined, the relationship between the SOREMP and Dream and NREMP and Nothing was reaffirmed.

These findings support our hypothesis that Dreams produced during REM sleep are associated with the appearance of REM stage itself, when pre-sleep processes are controlled. Thus SOREMP Dreams appear to relate to the mechanisms that are active in the production and maintenance of REM sleep (amount of stage REM). In contrast, NREMP Dreams in this study appear to be associated with brief arousal intrusions (stage W, body movements) out of sleep rather than with the sleep process itself (stage 1 or 2). Thus, Dreams seem to be produced by different mechanisms in SOREMP and NREMP sleep.

REM sleep is known for its specificity in terms of phenomenological and neurobiological aspects (for review see Hobson et al. 1998, 2000; Gottesman 1999). In addition, current evidence concerning REM sleep seems to support our findings that REM mechanisms contribute to dreaming. For instance, functional specificity during REM sleep such as memory consolidation and learning has been suggested (De Koninck et al. 1989; Stickgold et al. 1999; Karni et al. 1994; Plihal and Born 1997; Smith 1995). Moreover, De Koninck et al. (1989) showed that for participants who made significant progress in learning French, their learning experiences were incorporated into their dreams. Llinas and Ribary (1993) found that 40-Hz gamma EEG activity was not reset by sensory stimuli during REM sleep but was when participants were awake. Based on their results, they suggest that specific loops give the content of cognition, and a non-specific loop gives the temporal binding required for the unity of cognitive experience.

In addition, PET studies have shown anatomical specificity during REM sleep such as decreased activity in the dorsolateral prefrontal cortex (Braun et al. 1998; Maquet et al. 1996). Thus, Braun et al. (1998) have proposed that during REM sleep, the visual system operates as a closed loop, without input from the primary visual cortex and without integration of visual information in higher association areas of the frontal cortex. In other words, visual stimuli may be generated internally during REM sleep, rather than being information processed from external sources. In addition, Madeson et al. (1991) reported decreased blood flow in the inferior frontal cortex in subjects that were awakened from REM sleep during dreaming. They suggested that the decrease in activation levels in the frontal cortex might reflect ‘the poor temporal organization and bizarreness often experienced in dreams’.

Another possible explanation for the eight NREMP Dreams obtained in this study could be the intrusion of the covert REM process during NREM sleep (Nielsen 2000). These NREMP dreams might have led to SOREMP if participants had been sleeping without interruption for more than 5 min, as SOREMPs are defined as REM periods with a latency of up to 25 min (Miyasita et al. 1989a; Bes et al. 1996). Sasaki et al. (2000) used the Sleep Interruption Technique in the 2nd and 4th cycle, although participants were not awakened until REM sleep appeared. In their results obtained from the 4th cycle (same as our protocol) among 32 sleep episodes after interruption, the frequency of REM latencies that were less than 5 min was 15 (46.8%), while the frequency of latencies between 5 and 25 min was 13 (40.6%), and those more than 40 min was 4 (12.5%). Considering that 42 SOREMP episodes (39.2%) in our study had a REM latency of less than 5 min due to the nature of the protocol, among the remaining 65 NREMP episodes, there is a quite high possibility that most of these episodes might have been SOREMP with 5–25 min REM latency. Hence, we speculate that these 8 NREMP episodes reported as Dreams might have contained ‘covert’ SOREMP.

A further possible factor that may have caused NREMP Dreams is the contribution of some kind of arousal process during NREM sleep. Micro-arousals during sleep occur as often as 4–26 per hr depending on the definition of arousal (Mathur and Douglas 1995). Participants are often unaware of these arousals even though their EEGs clearly show arousal during sleep (Ogilvie et al. 1989; Ogilvie 2000). Moreover, external stimuli have been shown to influence mentation during NREM sleep but not REM sleep (Castaldo and Shevrin 1970; Conduit et al. 1997). Considering these findings and the strong relationship between NREMP Dreams and awakening in our results, we postulate that arousal processes might be related to Dream production during NREM sleep. That is, people might incorporate some information from external surroundings into their memory during brief arousals, and later amend or reconstruct these mentations as ‘Dreams’ as if they had been experienced during sleep. Thus, mechanisms of NREMP Dreams appear to be different from those of SOREMP Dreams even though they are regarded as the same Dreams in phenomenological terms. All in all, these data provide support for our hypothesis that there are different physiological mechanisms underlying Dream production in NREMP and SOREMP episodes.

Qualitative aspects of SOREMP/NREMP dreaming

In terms of the quality of dreams, our results indicated different properties within NREMP and SOREMP episodes. Previous research has suggested that there are no significant differences in qualitative aspects of dreaming between NREM and REM sleep when narrative length is controlled (Antrobus 1983; Cavallero et al. 1990; Foulkes and Schmidt 1983). However, Hunt et al. (1993) showed evidence that discredited the validity of the control of narrative length. Their results showed that describing a bizarre pictorial stimulus led to the use of more words than describing a mundane stimulus. They also suggested that speech fluency might affect the content of dreams, yet eliminating its effect may also eliminate real properties of a dream. A simple rating scale such as the DP scale would be able to reconcile both aspects in terms of the global and the formal levels of analysis in empirical dream research as Hobson and Stickgold (1994) have suggested. The DP scale focuses on the participant’s entire global experience of the dream as an image and is not dependent on speech fluency. At the same time, it is capable of capturing the relative structural differences of dreams among experimental conditions.

Our results using the DP scale showed that SOREMP Dreams were characterized as brighter, more cheerful, more comfortable, more likable (Evaluation), clearer, more explicit, more memorable and more vivid (Impression) as well as noisier, more dynamic, and more bustling (Activity) compared with NREMP dreams. These differences between SOREMP and NREMP suggest that the global properties of Dreams are different between these two types of sleep even if the participant perceived them both as Dreams. This also supports our hypothesis that there are different physiological mechanisms underlying dream production in NREMP and SOREMP episodes.


This comparison of Dreams obtained in SOREMP and NREMP using the Sleep Interruption Technique suggested that dreams were different in terms of both quantitative and qualitative aspects. More Dreams were obtained in SOREMP than NREMP. Furthermore, the appearance of SOREMP Dreams was related to an increased amount of stage REM, while the appearance of NREMP Dreams was related to an increased amount of arousal factors during NREM sleep. Dream properties were also different between SOREMP and NREMP. In attempting to generalize these findings obtained with the Sleep Interruption Technique to typical REM and NREM sleep, the following conclusion can be made: (1) dreams obtained from REM and NREM seem to differ in both quantitative and qualitative aspects when the effects of previous sleep stage are removed; and (2) although ‘Dreaming’ may occur during both REM and NREM periods, as previous researchers have suggested (Antrobus 1983; Antrobus et al. 1995; Cavallero et al. 1992; Cavallero et al. 1990; Foulkes 1993; Foulkes and Schmidt 1983; Rosenlicht et al. 1994), these dreams are likely to be produced by different mechanisms during REM and NREM sleep.


We thank R. D. Ogilvie, T. Murphy, W. Veenhof, P. Pailing, A. J. Wintink, S. J. Segalowitz, N. DeCourville, K. Fukuda, P. Moore, R, Stickgold, and E. Pace-Schott as well as the anonymous referees for their useful comments and suggestions with this manuscript. We also thank the first author’s mentor, A. Miyasita, who developed the Sleep Interruption Technique but sadly passed away in 1996. The first author was supported by the Japan Society for the Promotion of Science during this study. This work was supported by a Grant-in-Aid for Scientific Research from the Ministry of Education and Culture, Japan.