Seventeen amnesics, including patients with Korsakoff's disease, post-encephalitic amnesia and amnesia caused by rupture of an anterior communicating artery aneurysm, were compared with 17 matched control subjects on a task in which 16 nameable shapes were placed on different squares of a 49-square grid. One version of the task tapped free recall and recognition of the shapes and a second version tapped three forms of spatial memory. The patients were tested after more learning opportunity and shorter delays than were the controls so as to match their recognition levels. Under these conditions, the amnesics performed worse than the controls on free recall, location-to-target memory, target-to-location memory and possibly on non-associative spatial memory although this was less impaired than target-to-location memory. Each of the main aetiological subgroups showed these disproportionate deficits to an apparently similar degree. The disproportionate free recall deficit was unrelated to the spatial memory deficits and overall severity of amnesia, and was associated with ageing and signs of frontal lobe dysfunction. The disproportionate spatial memory deficits were unrelated to frontal lobe dysfunction, but the target-to-location memory measure was associated with impairments of recognition and recall of target material. The results are broadly consistent with the context-memory deficit hypothesis of amnesia.