In 1956 Bruner, Goodnow, and Austin proposed that there are four "ideal" strategies that can be employed to solve a selection concept-learning-task: simultaneous scanning, successive scanning, conservative focusing, .and focus gambling. They suggested that any particular problem solver only approximates an ideal strategy, and they presented methods for identifying the selection strategy most nearly approximated. In the intervening years, these methods have frequently been criticized for their ambiguity, and various alternative methods have been suggested.This article argues that the strategy identification procedures proposed by Bruner et al. and by their critics have been incomplete because little if any evidence has been presented to demonstrate that the procedures are valid.It is argued that a strategy identification procedure must meet five tests in order to demonstrate validity. First, the distribution of the primary data, such as trials to solution, must show evidence of bimodality or discontinuity, which would indicate that not all observations are random departures from a single solution method. Second, a procedure must be specified for rigorously and reliably assigning protocols to strategy classes. Third, it must be shown that subjects show a pronounced tendency to employ the same strategy when solving repeated problems of the same type. Fourth, the performance of the strategy groups must relate in a sensible manner to other task variables, such as latency, that were not used in strategy identification. Lastly, the strategy classifications must show an ability to predict relevant nontask variables, such as intelligence, and to predict performance on other tasks.The steps to be performed in demonstrating validity are illustrated with data gathered from 130 individuals, all of whom solved several concept-learning problems, solved two other computer-administered problem-solving tasks, took the Concept Mastery Test and Raven's Progressive Matrices test, and provided access to their Scholastic Aptitude Test scores. It is shown that several strategies (which are distinguishable on stylistic rather than efficiency measures) were employed to solve the concept-learning task, that they can be rigorously identified, and that the strategy employed on this task is predictive of performance on the other tasks. It is further shown that although the strategy groups differ in intelligence, the strategy classification provides useful predictive power beyond what can be predicted by intelligence alone.