In a previous study (Witte, 1959), a trace-fluctuation model derived from statistical learning theory was applied to choice behavior under sequentially dependent reward schedules. An adequate test of the theory was hampered by the large number of 5s with position preferences. The present study attempts to test the theory further by using a procedure designed to yield within-5 comparisons, as well as to minimize position preferences.When reinforcement is independent of 5's responses and occurs on every trial, the trace-fluctuation model generates the well-known "probability-matching" prediction for unconditional response probability. When, in addition, trials are massed, the model predicts that conditional response probability will be a direct function of the sequential dependencies between reinforcements. More specifically, the asymptotic probability of a response similar to the one reinforced on the previous trial, henceforth called a "repetition" response, following Anderson's (1960) usage, is [1]