A new subset selection algorithm is proposed, based on artificial intelligence search techniques. It selects the globally best subset with the optimal size from an element set without exhaustive enumeration. Existing procedures either do not guarantee the optimality of the selected subset or do not determine optimal subset size. Exhaustive searches, on the other hand, are generally computationally unfeasible. The adoptability of the present algorithm is proved. The experiments, in the applications of the algorithm to clustering and feature selection, show its effectiveness.