Previous studies have shown that self-generated information is better remembered than information that has been read passively. To further examine this subsequent memory effect, we investigated the effect of five different linguistic relationships on memory encoding. Ninety subjects were administered 60 paired associates during an encoding condition: 30 of the second words from each pair were to be read aloud and 30 were to be self-generated from clues as to the correct word. Word pairs were composed of five linguistic relationships: category, rhyme, opposite, synonym, and association. Subsequently, subjects were presented with the words that were read or generated in a forced recognition memory task. Overall, reading accuracy was higher than generation accuracy during the encoding phase (all P < 0.001). During the recognition phase, subjects' performance was better on the generate than on the read conditions for opposite, synonym, category, and association relationships (all P < 0.05), with no difference in the rhyme relationship. These results confirm previous findings that self-generated information is better remembered than read information and suggest that this advantage may be mediated by using opposite, synonym, category, and association relationships, while rhyme relationship may not extend such an advantage. These findings may have implications for future studies of memory interventions in healthy controls and subjects with cognitive impairments.