This study demonstrates and discusses a new procedure for performing item analysis which utilizes multiple discriminant analysis to establish efficiently and effectively an index of item validity. Application of this statistical technique to data derived from an attitude survey of three groups of students enrolled in technical training courses yielded the following results: It disclosed those stimulus items which were responsive enough to discriminate among criterion groups; it partitioned the total discriminatory power of the items into two homogeneous components; it yielded data for arriving at a special weighting scheme for scoring the final attitude form; and it located the positions of the criterion groups relative to the two orthogonal dimensions of the attitude universe. (Author)