This document compares the use of decision tree approaches to IRT-based CAT (computerized adaptive testing) for adaptive item selection and score estimation. Decision trees use predictor variables to partition a sample into increasingly homogeneous subgroups, represented as nodes in a tree structure. The study used decision tree algorithms and IRT modeling to select items and estimate scores on substance abuse scales. It found that decision trees were more efficient initially but that CAT outperformed decision trees in later stages of administration and had higher sensitivity to group differences. The authors conclude that combining decision trees with CAT may provide advantages.