Although data on the prevalence of injection drug use are an essential prerequisite for estimating the number of individuals infected with the human immunodeficiency virus (HIV), there have been few attempts to utilize statistical methods of population estimation based on multiple data sources. Data on 3,670 cases (2,866 individuals) were obtained from the HIV test register, drug treatment agencies, police records, and needle and syringe exchanges in Glasgow, Scotland, in 1990. Log-linear analysis was used to model the number of individuals in each of the sources. The model incorporating dependency among the three health care agencies (HIV test, drug treatment, and needle exchange) and independence of the police sample fitted the data well, with a residual chi 2 value of 2.9 (6 df). The expected value of the missing cell corresponding to absence from all four samples was 5,628, yielding an overall estimate of 8,494 injectors (95% confidence interval (CI) 7,491-9,721), for a prevalence rate of 1.35% for people aged 15-55 years in Glasgow during 1990. The high ratio of known to unknown injectors (1:2) resulted from the extensive coverage of known injectors and the relatively high level of overlap between the combined health care agency sample and the police sample. While further analysis demonstrated that the probability of appearing in the four samples varied by age and sex, heterogeneity in the population did not affect the choice of model or substantially alter the estimates for the total number of unknown injectors. A concurrent study of a community-wide sample of 503 injectors resulted in an HIV prevalence rate of 1.1% (95% CI 0.4-2.5%). The results of these studies were combined to produce a further estimate of 93 HIV-infected current injectors in Glasgow (95% CI 33-214).