The study found that the types and rates of classification errors were dependent on both the choice of the cutoff score and the performance measure to be predicted. At the 16 or higher cutoff score, African-Americans were substantially more likely to be overclassified than were either Caucasians or Hispanics when measured by program outcome. Overclassification differences were smaller for disciplinary incidents, although they followed the same pattern. At the 25 or higher cutoff score, however, there was a smaller difference in overclassifications between African-Americans and Caucasians; both were overclassified somewhat more than were Hispanics for program outcome. African-Americans were significantly more likely to be underclassified than were Caucasians or Hispanics when predicting disciplinary incidents. Although types of classification errors varied by cutoff score and performance measure, there was an overall trend toward more classification errors for African-Americans than for Caucasians and Hispanics. The findings suggest that correctional facilities that use classification instruments should conduct local validation studies to ensure that the current use of their instrument does not produce classification errors for different racial or ethnic groups, which results in inappropriate treatment and security measures. The study samples consisted of 532 male residents who entered a large Federal residential work-release facility in a midwestern metropolitan area after January 1, 2002, and were released by December 31, 2002. The residents' risk classification based on their LSI-R score was the independent variable. In order to test for classification errors, the study measured program outcomes for the sample and in-program disciplinary incidents. 4 tables and 22 references