Neurophysiological studies in monkeys suggest selective representation of behaviourally relevant information in working memory. So far, no behavioural evidence for this has been reported for humans. Here, we investigated the role of behavioural relevance for access to human visuospatial working memory by using delayed oculomotor response tasks. Subjects were presented two successive visual cues in different and unpredictable locations while fixating on a central fixation point. After a delay, an unpredictable auditory signal (one beep or two beeps) sounded and the central fixation point was turned off, initiating the oculomotor response (i.e. memory-guided saccade) phase. Two groups of 10 subjects each were studied in two conditions: in the ‘relevant’ condition, subjects were instructed to memorize both visual cues and to move the eyes to the remembered position of the first cue (one beep) or the second cue (two beeps). The same stimuli were used in the ‘irrelevant’ condition, but subjects were instructed to memorize and move the eyes to the position of the first cue only, regardless of the second cue and the auditory signal. In the ‘relevant’ condition, we found a significant increase in errors of memory-guided saccades to the first cue, when the second cue was located between central fixation point and first cue. This spatially selective interference effect disappeared in the ‘irrelevant’ condition, despite identical stimuli. On a behavioural level, these results show for the first time the significance of behavioural relevance for access to human spatial working memory. These findings complement recent single-neuron studies in monkeys, showing that the neuronal substrates of working memory selectively represent behaviourally relevant perceptual information.