The main advantage of tree classifiers is to provide rules that are simple in form and are easily interpretable. Since decision tree is a top-down algorithm using a divide and conquer induction process, there is a risk of reaching a local optimal solution. This paper proposes a procedure of optimally determining the splitting variables and their thresholds for a decision tree using an adaptive particle swarm optimization. The proposed method consists of three phases – tree construction, threshold optimization and rule simplification. To validate the proposed algorithm, several artificial and real datasets are used. We compare our results with the original CART results and show that the proposed method is promising for improving prediction accuracy.