Twelve Carneaux pigeons were divided into three groups and trained on zero-delay matching-to-sample, fixed ratio matching-to-sample or zero -delay non-.natching. Reinforcing only every third correct match with grain was found to substantially slow acquisition. Learning matching or non-matching with red and green stimuli did not produce generalized transfer nor did the transfer task interfere with performance on the original problem. Re-pairing the stimuli so as to change the odd comparison stimulus was shown not to affect matching performance but to cause a decrement in non-matching in two out of three cases. Interpolation of a one second delay between the offset of the standard stimulus and the onset of the comparisons caused all animals to drop to chance performance, from which they never improved. The results are interpreted in terms of the coding hypothesis of Cumming ert. al. (1965).