All-or-none-acquisition in matching-to-sample and a test of two models.

Three pigeons learned to match three colors in a matching-to-sample procedure. The sample hue was presented on the center one of three keys, and the comparison hues were presented on the side keys. Forty responses on the center key produced the stimuli on the side keys and left the sample on the center key. A single response on the correct side key produced 3-sec access to grain, which was followed by a 25-sec intertrial interval. A correction procedure was employed when an error was committed. Before attaining asymptotic levels, there was no evidence of learning, responses were independent of the preceding response, and distributions of errors in four-trial blocks were binomial. Distributions of error runs, runs of various lengths, autocorrelations of errors of several lags, alternations of correct responses and errors, etc., were shown to fit Bower's (1961) all-or-none model better than a gradual learning model of Bush and Sternberg (1959). A transfer test employing a novel color showed only transitory degradation of performance.