Data were collected on all persons who agreed to participate in the program (n=787). Subjects were interviewed within several days of their arrival at the drug treatment facility. Only 418 completed the program; the others had their cases dropped or otherwise left the program. All drug-use and demographic information used in this analysis were obtained from the self-report interviews. The demographic information was checked against the same information reported to the program by clinical staff. Logistic regression analysis was used to identify significant predictors of program graduation. The research shows that demographic characteristics, such as race and education, as well as crack use, are the most useful defendant characteristics for predicting program graduation. Although policy makers are unable to change the educational levels of those entering their programs, they might focus on increasing the participants' education attainments to both improve program success and help ensure that participants will not be required to enter similar drug programs in the future. Program administrators might also consider developing various types of treatment curricula commensurate with the various levels of educational attainment among participants. Further, programs might develop special strategies for dealing with crack cocaine users, such as distinctive program curricula and individual treatment protocols. 2 tables and 49 references