Attempts to characterize voluntary behavior have been ongoing for thousands of years. We provide experimental evidence that judgments of volition are based upon distributions of responses in relation to obtained rewards. Participants watched as responses, said to be made by "actors," appeared on a computer screen. The participant's task was to estimate how well each actor represented the voluntary choices emitted by a real person. In actuality, all actors' responses were generated by algorithms based on Baum's (1979) generalized matching function. We systematically varied the exponent values (sensitivity parameter) of these algorithms: some actors "matched" response proportions to received reinforcer proportions, others "overmatched" (predominantly chose the highest-valued alternative), and yet others "undermatched" (chose relatively equally among the alternatives). In each of five experiments, we found that the "matching" actor's responses were judged most closely to approximate voluntary choice. We found also that judgments of high volition depended upon stochastic (or probabilistic) generation. Thus, stochastic responses that match reinforcer proportions best represent voluntary human choice. (Contains 3 tables, 3 footnotes, and 8 figures.)