AN EMPIRICAL STUDY WAS MADE OF THE ERROR FACTORS IN MULTIDIMENSIONAL SCALING (MDS) TO REFINE THE USE OF MDS FOR MORE EXPERT MANIPULATION OF SCALES USED IN EDUCATIONAL MEASUREMENT. THE PURPOSE OF THE RESEARCH WAS TO GENERATE TABLES OF THE SAMPLING DISTRIBUTIONS THAT ARE NECESSARY FOR DISCRIMINATING BETWEEN ERROR AND NONERROR MDS DIMENSIONS. THE TABLES DEVELOPED WERE USED TO TEST THE HYPOTHESIS THAT A GIVEN MDS FACTOR IS NO GREATER THAN ONE ATTRIBUTABLE TO A COMBINATION OF MEASUREMENT ERRORS. SEVERAL VARIABLES THAT COULD INFLUENCE SAMPLING DISTRIBUTIONS OF ERROR ROOTS WERE CONSIDERED. THIS RESEARCH INVOLVED THE SIMULATION OF JUDGEMENTS OF 15 OBJECTS BY 35 JUDGES ACCORDING TO THE MULTIDIMENSIONAL RANKING PROCEDURE OF KLINGBERG (1941). THE EFFECTS OF THE VARIABLES WERE CONSIDERED BY VARYING THE NUMBER OF JUDGES FROM 20 TO 80 AND BY INDEPENDENTLY VARYING THE NUMBER OF OBJECTS FROM 8 TO 15. COMPUTER PROGRAMS FOR THIS INVESTIGATION WERE BASED ON PROCEDURES OUTLINED BY TORGERSON (1958). THE STUDY LED TO THE CONCLUSIONS THAT (1) ONLY A SINGLE NONERROR DIMENSION COULD BE DESIGNATED WITH STATISTICAL CONFIDENCE, AND (2) ASCERTAINING THE NONERROR DIMENSIONALITY OF THE STRUCTURE COULD NOT BE ACCOMPLISHED WITH THESE STRUCTURE STATISTICS ALONE. (AH)