A broad range of baseline subject variables was evaluated to identify predictors of 7-month cocaine use for 160 lower socioeconomic cocaine dependent male veteran patients participating in either an intensive 1-month day hospital (DH; n=90) or a 1-month inpatient (INP; n=70) treatment program. The baseline measures included sociodemographic variables, the seven Addiction Severity Index composite scores, cocaine urine toxicology, craving, the SCL-90 total score, and lifetime psychiatric diagnoses. Since a proportion of subjects who reported no use at follow-up had positive urines, both liberal and conservative data estimation strategies were employed for subjects without urine toxicology data at follow-up who had reported no use (21% of subjects). Analyses were done separately for the DH and INP subjects. Under the conservative definition of cocaine abstinence/use, univariate correlations of predictor variables with 7-month cocaine use revealed no statistically significant relationships. Under the liberal definition of cocaine abstinence/use, only one variable, greater severity of alcohol problems at intake predicted cocaine abstinence at outcome. Because of the inability to predict treatment success, originally planned logistic regression analyses were not undertaken. The findings point to the difficulty of predicting long-term outcomes in cocaine dependent patients based on baseline information and to the importance of obtaining objective data on cocaine use.