The experiment concerns the specificity of transfer between prior study and a later performance measure of remembering, picture naming of previously seen and new pictures. Korsakoff's syndrome and normal subjects viewed pictures under two sets of instructions: integral, where target pictures were to be integrated with context pictures; and separable, where target pictures appeared alone. At test, subjects' naming latencies for targets which were either (a) old and in the same orientation as at study; (b) old but in a reversed orientation; or (c) new, showed that picture naming is sensitive to the unit of processing of the prior episode. When no active processing was required (separable instructions) Korsakoff subjects were as sensitive as normals to the prior physical stimulus. However when active integral processing was required Korsakoffs and normals showed differential sensitivity to the prior episode, consistent with the Korsakoffs' preference for passive rather than active processing.