A working model of depression implicating failure of the coordinated interactions of a distributed network of limbic-cortical pathways is proposed. Resting state patterns of regional glucose metabolism in idiopathic depressed patients, changes in metabolism with antidepressant treatment, and blood flow changes with induced sadness in healthy subjects were used to test and refine this hypothesis. Dorsal neocortical decreases and ventral paralimbic increases characterize both healthy sadness and depressive illness; concurrent inhibition of overactive paralimbic regions and normalization of hypofunctioning dorsal cortical sites characterize disease remission. Normal functioning of the rostral anterior cingulate, with its direct connections to these dorsal and ventral areas, is postulated to be additionally required for the observed reciprocal compensatory changes, since pretreatment metabolism in this region uniquely predicts antidepressant treatment response. This model is offered as an adaptable framework to facilitate continued integration of clinical imaging findings with complementary neuroanatomical, neurochemical, and electrophysiological studies in the investigation of the pathogenesis of affective disorders.