This study was concerned with whether percent-choice data in probability learning may include subclasses with different choice proportions. Percent choice of the positive alternative was significantly greater, even at asymptote, when the positive alternative was on the right. Choice latency was also recorded; there was a significant tendency to respond more rapidly when an asymmetrically located reinforcement signal was on the right or when it was on the side opposite the positive alternative. The percent-choice data did not show corresponding significant effects. These findings suggest that percent-choice measures may pool a variety of different ways of responding that are differentially affected by experimental variables.