Since the majority of persons with alcohol, drug abuse, and/or mental disorders (19%) of Americans during any 6-month period are seen exclusively within the general health sector, it is imperative to know the quality and quantity of mental health training for primary care residents. In this study, the five program training model types previously described--Consultation, Liaison, Bridge, Hybrid, Autonomous--are validated by a random sampling technique using a structured instrument to test eight hypotheses developed before data collection to preclude post hoc interpretations. Of 250 programs, 147 responded (60%): 67 Family Practice, 42 Primary Internal Medicine, and 38 Internal Medicine. Since all eight hypotheses were supported by the data, the construct validity of the program model types is significantly substantiated. Multiple discriminant analysis revealed that the relationships between twelve training program characteristics and the five program model types were such that the former could explain 57% of the variance in the latter and correctly classify 89% of the programs.