Seventy-two Ss inspected computer-generated random dot matrices under 3 size-of-matrix conditions. The method of constant stimuli was employed with Ss asked to judge whether the dots tended to chain together or be evenly spaced along the rows, a tendency controlled by the transition probability of a Markov process used in generating the matrix. Functions relating judgment latency to the proportion of transitions in the matrix were found to be shaped, and the accuracy functions derived were satisfactorily fitted by normal ogives. Both overall latency and accuracy of judgments increased when matrix size was increased, and interpretation of these effects in terms of a statistical model suggested that Ss were responding not to the proportion of transitions, but to some other aspect of the matrix correlated with it.