This study is based on a 'natural experiment' in which a cohort of heroin users was assessed at one unit, then referred on geographic grounds for treatment to one of two clinics--one orientated to long-term maintenance (Clinic 2, with 61 subjects), the other to time-limited treatment aimed at achieving abstinence from all drugs including methadone (Clinic 1, 141 subjects). The outcome measure was heroin use as measured by urine testing performed regularly at both clinics. Overall, 25% of urine tests from Clinic 1 were positive for heroin compared to 18% in Clinic 2. This difference reflected in part a high rate of heroin use during the period of mandatory withdrawal from treatment in clinic 1. Statistical models were developed to identify factors associated with heroin use. There was a strong association between methadone dose and heroin use; relative to a daily dose of 40 mg, a dose of 80 mg/day of methadone was less likely to be associated with a heroin-positive urine (OR 0.55, 95% CI [0.45, 0.68]). Average doses prescribed in Clinic 1 were lower, reflecting the clinic's orientation to abstinence. Adjusting for dose, and for the fact that certain individuals tend to use heroin heavily while others do not, there was no difference between the clinics in risk of heroin use during maintenance treatment. The higher rates of heroin use in the abstinence-orientated clinic were attributable to time-limited treatment and the use of lower doses of methadone. This finding confirms that in investigating the effects of treatment factors, the powerful influence of methadone dose needs to be taken into account.