As U.S. general internists play an increasing role in providing opioid maintenance therapy in practice offices, they are having to face the challenge of identifying patients who need specialized services especially at the outset of treatment. In methadone maintenance treatment, prognostic studies have failed to find robust predictors on the basis of single predictive variables. We hypothesize that a multivariable staging system will predict treatment outcome more accurately than single variables. We reviewed baseline and treatment data regarding 226 consecutive patients admitted to a methadone maintenance program in New Haven, Connecticut, from January 1, 1993 to March 28, 1994, and followed until December 1, 1996. The staging system was developed from the data on the first 112 patients, confirmed in the remaining 114 patients, and then applied to the entire cohort of 226 patients. Retention was the main outcome measure used in developing the staging system. The staging system was also validated as a predictor of illicit drug use during treatment and adverse discharge. In the staging system one point is scored for each of the following: use of more than two bags of heroin daily, previous prison term, previous period in reform school, and a history of diseases related to substance use, e.g., endocarditis, hepatitis, abscesses, and overdose. The total score classifies patients as Stage I (0 and 1 points), Stage II (2 points), or Stage III (3 and 4 points). This staging system was significantly associated with retention in a proportional-hazards model, and no other variable added any additional predictive influence. The specific stage was also found to be a significant predictor of adverse discharge. Although additional validation is necessary in other populations, we found the staging system to be a useful and simple way of identifying patients at risk for early attrition and adverse discharge.