The effects of imagery and semantic relatedness on cued retrieval of word pairs were examined in a functional imaging study of healthy volunteers. Subjects underwent 12 PET scans, preceded by the paced presentation of 12 paired associates. The associates were dichotomized into imageable and non-imageable groups. Within each group, the strength of semantic association between members of pairs was varied in an ordinal fashion. Subsequently, neural activity was measured while subjects were cued with the first item of each pair and required to recall the associated word. Recall of imageable words, when compared with non-imageable ones, was associated with activation of the precuneus, consistent with our hypothesis that this region is important in visual imagery at episodic retrieval. The reverse comparison, non-imageable versus imageable recall, was associated with activation of the left dorsolateral prefrontal cortex. Within both imageable and non-imageable groups, decreasing semantic association showed a corresponding increase in frontal activity bilaterally. One possible explanation is that of a practice-related effect, weaker-linked pairs having a greater number of pre-scan presentations. However, this explanation is incomplete as the most semantically distant, and most rehearsed, pairs (randomly linked) were associated with a reversal of this effect. This finding can be explained if frontal activity is associated with the difficulty of eliminating inappropriate responses at retrieval. For both randomly linked pairs and closely related pairs it is more likely that erroneous responses will be generated and, therefore, the work done to eliminate them will be greater. Our findings indicate that patterns of neural activity during cued recall depend upon the nature of the material and on the degree of association between the cue and the response.