We analyzed the electrical generators of CNV-type potentials obtained during a visual paired-associates memory task, using Lp norm minimization constrained by individual cortical anatomy. The task that we designed maximized the chances of verifying prefrontal cortex activity, by combining the anticipation of interesting, relatively arousing stimuli with the uncertain expectation based on memory recall. We tested a set of topographical hypotheses on the relation between activity in particular prefrontal domains and task conditions. The task consisted in two types of blocks: 1) memorization, when pairs of words, or abstract figures or positions on the screen were separated in time by 2,5 seconds @l-S2 stimuli); 2) execution with feedback for each trial, when subjects had to decide whether the upcoming pair (S2) of a given stimulus (Sl), just seen in the previous memorization block, formed or not a category match. Sl remained on the screen until a button press indicated the subject’s decision. Two and a half seconds after button press, S2 was presented, and 2,5 seconds after S2 onset, the feedback stimulus (S3) was presented. An interesting, relatively arousing visual stimulus indicated correct and an unpleasant sound indicated incorrect performance. We recorded the EEG using a 123-channel montage in 19 healthy subjects We used a realistic, three compartments model of the head, based on segmentation of individual TZweighed MRI sets and approximation by few milimeter side triangles. Approximately 20.000 points distributed over the cortical surface supported possible model equivalent current dipoles. The average slow potentials were then transformed into signal-to-noise ratios, and the intracranial electrical generators modeled as distributed current sources. For solving the inverse problem (finding the best fitting current distributions that explain the surface data), we applied a minimization algorithm to the Lp (p= 1.2) norms of data and model terms. The generators of the feedback anticipation CNVs consisted in a set of cortical association areas, including prefrontal Brodmann areas 9 and 10 in all subjects. We also obtained preliminary evidence, by analyzing potentials averaged across subjects, for task-specific (verbal, spatial and pictorial) topography of the memorization CNVs. The generators of such group average potentials were also a collection of cortical areas, with some areas appearing to be specifically active during the spatial subtask (left areas 9,10, lateral 6,5,39 and right 19), as opposed to some areas active during the verbal subtask (right 9,22 and 7). The pictorial subtask seemed to combine activity from the other two subtasks. We propose the use of current density reconstruction of the generators of slow potentials as a method for evaluating the functional integrity of cortical association areas in neuropsychiatric clinical conditions. We have started to use such method in a group of schizophrenic patients, and found preliminary evidence, using group averaged potentials, for important abnormalities in signalto-noise ratios and in the topography of all CNVs in the schizophrenic group. The global field power of the memorization CNVs was enhanced in the three subtasks. Finally, preliminary source reconstruction, by the projection of group potentials onto representative head models, indicated a larger degree of generator dispersion in the group of patients.