The American College Testing (ACT) Program offers research services through which colleges can predict the freshman grades of their future students. This paper describes research done to establish a minimum sample size requirement for calculating least-squares prediction equations for college freshman grade average. Prediction equations were developed from the 1974-80 freshman grade data at all colleges participating in the ACT Research Services with between 70 and 100 freshmen. Separate subgroup equations were also developed for the males and females at each college. The prediction equations were then cross-validated against the grades of the 1981-82 freshmen at each college. Results confirmed that total group predictions based on 70 or more students have the same accuracy as predictions based on large samples. Moreover, the results from separate-sex prediction equations lent further support to the idea that a base sample size as low as 50 would be satisfactory. (Author/BW)