The factors which prevent the a priori prediction of the impact of training on recidivism are discussed. In contrast to most prior research which measured the training experience with a single trainee-control variable, this study presents four sets of estimates based on a logistic probability model which progressively disaggregates the training characteristics. This approach permits the identification of a trainee subgroup for which the probability of recidivism was significantly reduced by enrollment in a vocational training program. Tables, 12 notes, and 17 references are included. (Author abstract modified)