This paper examines issues that must be considered when evaluating generalized identity matching to sample. We suggest that (a) stimuli displayed together on tests of generalized identity matching should have comparable experimental histories, and (b) identity matching procedures should be differentiated as either “conditional” or “nonconditional,” depending upon whether or not conditional discrimination is required. Although either type of procedure might be adequate for demonstrating a “matching concept,” we suggest that only generalized conditional identity matching provides an adequate test for the reflexive property of conditional relations.