The technique of nonmetric multidimensional scaling (MDS) was applied to real item response data obtained from a multiple-choice achievement test of unknown dimensionality. The goal was to classify the 50 items into the various subtests from which they were drawn originally, the latter being unknown to the investigator. Issues addressed in the research included dimensionality, choice of item proximity measures, and appropriateness of the MDS model for analyzing dichotomous item response data. Three coefficients were chosen to form proximity matrices that reflected the associations of each item with each of the other items. These matrices then served as input to both the ALSCAL and MINISSA computer programs for MDS analysis. A three dimensional solution was found to be optimal, based on stress, the multiple correlation coefficient, and coefficient of alienation values. Both cluster analysis and regional (facet theory) analysis of the spatial configuration were used to interpret the results. (Author)