People have generally been considered poor at probabilistic reasoning, producing subjective probability estimates that far from accord to normative rules. Features of the typical probabilistic reasoning task, however, make strong conclusions difficult. The present study, therefore, combines research on probabilistic reasoning with research on category learning where participants learn base rates and likelihoods in a category-learning task. Later they produce estimates of posterior probability based on the learnt probabilities. The results show that our participants can produce subjective probability estimates that are well calibrated against the normative Bayesian probability and are sensitive to base rates. Further, they have accurate knowledge of both base rate and means of the categories encountered during learning. This indicates that under some conditions people might be better at probabilistic reasoning than what could be expected from previous research.