The goal of this study was to explore in patients with Alzheimer's disease (AD) the brain correlates of free and cued recall performance using an adaptation of the procedure developed by Grober and Buschke (1987). This procedure, which ensures semantic processing and coordinates encoding and retrieval, has been shown to be very sensitive to an early diagnosis of AD. Statistical parametric mapping (SPM 99) was used to establish clinicometabolic correlations between performance at free and cued verbal recall and resting brain metabolism in 31 patients with AD. Results showed that patient's score on free recall correlated with metabolic activity in right frontal regions (BA 10 and BA 45), suggesting that performance reflected a strategic retrieval attempt. Poor retrieval performance was tentatively attributed to a loss of functional correlation between frontal and medial temporal regions in patients with AD compared with elderly controls. Performance on cued recall was correlated to residual metabolic activity in bilateral parahippocampal regions (BA 36), suggesting that performance reflected retrieval of semantic associations, without recollection in AD. In conclusion, this study demonstrates that the diagnostic sensitivity for Alzheimer's disease of the cued recall performance in the Grober and Buschke procedure (1987) depends on the activity of parahippocampal regions, one of the earliest targets of the disease. Moreover, the results suggest that the poor performance of patients with AD during free and cued recall is related to a decreased connectivity between parahippocampal regions and frontal areas.