A total of 10,010 clients were admitted to 96 programs in 11 cities that participated in the DATOS project during 1991-93. They included 2,774 admissions to 21 long-term residential programs, 3,122 admissions to 14 short-term inpatient programs, 2,574 admissions to 32 outpatient drug-free programs, and 1,540 admissions to 29 outpatient methadone treatment programs. Only programs that contained 50 or more clients who completed intake assessments were selected. Concurrent with length of stay in treatment, a set of basic client demographic and treatment- history measures was examined, along with variables that represented cocaine-crack dependence, HIV-AIDS risk factors, criminal justice involvement, and psychological dysfunction. DATOS findings show that stays of 3 months or longer in drug abuse treatment generally predicted better follow-up outcomes. Median lengths of stay were 3 months for clients in long-term residential and outpatient drug-free treatments and 1 year for clients in outpatient methadone treatment; however, individual programs within each of these modalities differed widely in how long they kept their clients in treatment, as well as their service delivery. Programs that treated individuals with heavier cocaine and alcohol use and more psychological dysfunction usually had shorter retention rates. Nonetheless, even after statistically controlling for these client differences, some programs were more effective than others in engaging and retaining clients. The implications of these and similar findings in drug treatment research conducted over the past 30 years are that programs should focus on increasing client retention rates so as to improve their effectiveness. Three months of treatment has empirical support as a minimum length of stay that can be recommended, and a year or more is usually required for opioid addicts in outpatient methadone programs. 3 tables and 41 references