In a same-different RT task, either an item stored for 3 sec. or a current visual stimulus could match a test probe. In 5 experiments, 100 undergraduates were faster in pressing the "same" lever when the stored item matched the probe, and this difference increased with trials. The effect was independent of anticipatory strategies and invariant over the nameable and name-ambiguous figures and color patches used as stimuli. When Ss overtly named the stored item before using it as a basis for making a memory match, RTs were inhibited. No facilitation of memory-match RTs was observed when Ss used the "real" name of a stimulus, e.g., "A" for an A, over using an arbitrarily assigned name, e.g., "A" for a nonsense figure. No difference was observed between memory-match RTs based on names and those which could be based on physical identity. Data provide little support for the notion that an item in short-term memory is necessarily represented as a name, or coded in some fashion which is dependent on nominal identity, and do not support the assumption that 2 stimuli must be named before they can be compared. Consideration is given to the possibility that the initial encoding of a stimulus may take time, somewhat less for colors than for figures, and that the major effect of practice is of an iterated recoding and simplification of anonymous mnemonic stimulus representation. (18 ref.) (PsycINFO Database Record (c) 2006 APA, all rights reserved)