The two experiments reported in this article examined recognition of simple and complex olfactory stimuli. In Experiment 1, three groups of control rats were trained to criterion without delay then, tested with delays on a continuous delayed nonmatching-to-sample (cDNMS) task using one of three kinds of odor pairs: replicates of the same odor (S), different odors that were each common with another pair (O), different odors with no overlap between pairs (NO). Results showed that initial learning and performance with delays were both poorer for O pairs than for S and NO pairs. Experiment 2 used a within-subject design to study the effects of fornix transection on recognition of the same three kinds of odor pairs as those described for Experiment 1. Sham-operated controls and rats (SH) with fornix transection were trained to criterion prior to the test with delays first on S, then on O and, finally, on NO pairs. During training, numbers of sessions to criterion did not differ in lesioned and SH rats on any of the three kinds of pairs. During testing, the level of performance was delay-dependent in both groups. However, lesioned rats were significantly impaired when tested with S and O pairs, but did not differ from sham-operated controls when tested with NO pairs. This selective impairment can be interpreted as evidence that fornix lesions impair recognition memory of stimuli that provide few and/or confusing retrieval cues. It might also suggest that postlesional performance on DNMS procedures depends on task difficulty.