The present study sought to clarify the influence of frequency and probability estimation on subsequent predictive choice performance. The experimental design involved a manipulation of quality of prior estimations via combinations of instructional set and response requirement. The results demonstrated that frequency judgments were consistently superior to probability judgments regardless of initial frequency or probability set. However, accuracy of predictive choices did not directly reflect the subjects’ estimation-performance. All experimental groups made better choices than a control group which had no set or estimation requirement; however, the experimental groups did not differ among themselves in choice performance. The latter finding suggests that estimation enhanced choice accuracy by cuing information that was accumulated in a similar fashion by all experimental groups: the record of event frequency. Some possible determinants of predictive choice for future practical and theoretical consideration are discussed.