The use of decision tree models for predicting activity-travel choice is receiving increasing attention, but raises two related problems that are considered in this study. First, commonly used deterministic action-assignment rules should be replaced by probabilistic action-assignment rules. We develop such probabilistic rules for both discrete and continuous choice problems. Second, common goodness-of-fit measures such as the hit-ratio need to be replaced by likelihood measures. In this paper, we develop and empirically illustrate the interrelated methods and measures. The findings suggest that the new measures add information to existing statistics for discrete as well as the continuous choice.