Pigeons trained initially to match element samples and element comparisons demonstrate reduced matching accuracy on trials in which the sample is a compound stimulus. Two interpretations of this phenomenon, the shared attention account and the coding decrement account, were assessed in three experiments. Experiment 1 obtained the typical outcome of reduced matching accuracy to compound samples following training to match to element samples. However, in a second group trained initially to match to compound samples, the opposite result was obtained: Birds matched more accurately on compound sample trials than on element sample trials. In Experiment 2 sample duration was manipulated, and in Experiment 3 birds were tested on simultaneous and 0-s delayed matching trials. Neither manipulation influenced the magnitude of the element-compound difference in either group. It was concluded that the coding decrement interpretation best accounts for these and other findings on matching to element and compound samples in pigeons. This account holds that (a) pigeons do not decompose compound samples; (b) presentation of a familiar sample activates a code; and (c) a novel sample which is sufficiently similar to a sample which activates a code will also tend to activate that code.