The achievement goal theory has been widely used as a framework to explain and to understand student motivation and related cognitive, affective, and behavioral outcomes in physical education. Classifying the participants into the groups based on their achievement goals is a critical step in applying the achievement goal theory to evaluate the cognitive, affective, and behavioral outcomes. In literature, k-means cluster analysis, hierarchical cluster analysis and a two-stage cluster analysis have been used for the classification purpose. However, different clustering methods could generate different groups with the same data set. The literature review shows that little research in our field was to evaluate the validity of the cluster analysis methods based on their weaknesses and strengths. Consequently, a procedure for appropriately classifying elementary school children based on the achievement goals is not available. The purpose of this study was to investigate how employing the different clustering methods would affect the membership of the achievement goal groups in elementary physical education. This project was intended to have a better understanding of the clustering techniques and to provide an appropriate classification procedure to classify students based on the achievement goals. Participants included 533 (285 boys and 248 girls) fifth graders who participated in a motivation study. The data of the Students' task- and ego-involved goals were used for this investigation. Students' task- and ego-involved goals were assessed using a 12-item questionnaire adapted from the Task and Ego Orientation in Sport Questionnaire (TEOSQ; Duda, 1992). No special pattern was noticed in the missing data (32 cases) on the task- and ego-involved goals. Thus, 498 participants' data were used in the three different cluster analyses after the missing data were deleted. The classification results of the students with the task- and ego-involved goal scores in physical education class were compared among the three clustering methods. The results indicated that different clustering methods yield different clustering solutions. There is no simple statistical test available to test either for the existence of clusters or the number of clusters using a single k-means and hierarchical clustering techniques. It is confirmed that a better approach is to use the hierarchical clustering for a preliminary analysis, and to use a k-means clustering with the cluster centers derived from the hierarchical clustering as the starting point for the final cluster solution. In spite of this, the cluster solution should be carefully validated in terms of internal validity and external validity.Keyword(s): physical activity, research