This study examined the usefulness of baseline cocaine urine toxicology results and self-reported days of cocaine use in predicting treatment response in cocaine- and opioid-dependent subjects. Ninety-nine male and 52 female subjects, maintained on buprenorphine, participated in a 24-week, randomized, double-blind, four-cell trial that evaluated desipramine (150 mg/d) or placebo plus contingency management or a noncontingent voucher control. Out of 151, 102 (67%) subjects had cocaine-positive and 49 (32%) cocaine-negative urines at the beginning of treatment. For the previous 30 days before study participation, 91 (60%) subjects reported using cocaine 15 or less days (low baseline cocaine use) and 60 (40%) subjects reported more than 15 days (high baseline cocaine use). By using the treatment effectiveness score (TES) as the outcome measure, a negative urine for cocaine at baseline predicted a better outcome during a 24-week trial for cocaine and opioid use. There also was a significant interaction between baseline cocaine urine results and desipramine response with the urine cocaine-negative group showing greater desipramine response than placebo for opioid and cocaine use. Self-reported cocaine use at baseline did not show significant predictive power for TES scores during the clinical trial. These results suggest that baseline cocaine urine results should be considered as stratifying variables in clinical trials for cocaine dependence.