The use of a combination of traditional cognitive admissions criteria and noncognitive variables in predicting retention of college students is discussed. The focus is on improving prediction by analyzing various subgroups separately rather than using the entire population as a sample. The subjects were freshmen in the fall semester of 1978 at a predominantly black university. Retention status was studied for a 5-year period for two subgroups: remedial versus regularly admitted students, and males versus females. Data were collected on race; sex; retention status; Scholastic Aptitude Test (SAT) scores; high school class rank; and noncognitive data, including motivational variables (e.g., extent of studying, ambition, satisfaction with grades). Higher multiple correlations between retention status and the various combinations of predictors were found for remedial students 4 of the 5 years after entry and for males as opposed to females in the fourth and fifth year after matriculation. The noncognitive data were collected during the first month of the freshman year; however, these variables could be studied prior to the admissions decision. (Author/SW)