Forced-choice tests of recognition have become the favoured behavioural method for the assessment of models of amnesia in nonhuman primates, yet the profile of deficits shown by human amnesic subjects remains uncertain. The present study explored the performance of 12 amnesic subjects on two delayed matching-to-sample tasks. Experiment 1, which used retention delays of between 2 and 60 sec, confirmed that amnesia impairs such tasks, even when there is only one item to be remembered. The results also highlighted the need to match levels of performance before the effects of delay can be interpreted. In Experiment 2 care was taken to eliminate ceiling effects and to match the subjects at the shortest delay (3 sec). This was achieved by giving the control subjects harder versions of the same task. The amnesic subjects still showed a faster rate of forgetting for abstract patterns, indicating that this is a genuine feature of amnesia. In contrast, the amnesic subjects' performance on a spatial matching-to-sample task was not differentially affected by delays of up to 40 sec. There was no evidence that the amnesic subjects were disproportionately impaired on this spatial task, nor could the different aetiological groups be distinguished by their patterns of DMS performance.