Reported are the results of a study designed to investigate and compare four cluster analytic procedures as potential methods for the analysis of educational data. A secondary objective was to determine whether or not there was some underlying multidimensional structure to a set of mathematics achievement data. The four clustering procedures (Ball and Hall's ISODATA, Johnson's HICLUS, Friedman and Rubin's iterative procedure, Singleton and Kantz's iterative procedure) were compared by applying them to a data set from the National Longitudinal Study of Mathematical Abilities of SMSG. The clustering variables were scales which described the characteristics of thirty junior high schools and their communities. The four clustering techniques produced very similar sets of clusters, and from all indications three or four clusters seem appropriate for clustering the mathematics achievement data. It was found that the students' mathematics achievement across clusters was not the same after adjustments were made for differences in aptitude and initial understanding of mathematical concepts. It was concluded that the differences in achievement were due at least in part to the effect of the particular school on the student. (Author/RS)