The discussion's goal is to highlight important features and limitations of current methods for estimating drug user populations, rather than to provide a comprehensive review. Definitions of such terms as prevalence, drug use, and population are considered, followed by a discussion of some basic aspects of prevalence estimation. An important consideration is that the choice of estimation methods must differ from drug to drug, since the likelihood that drug use will become a matter of record or be hidden from view differs greatly from one drug to another. In addition, a statistical model must be used to indicate how the data were generated and to explain how the prevalence estimates are produced. Moreover, confidence bands must be placed upon estimates of prevalence to avoid misleading anyone about the accuracy of the estimates. Furthermore, prevalence estimates may be made at various levels of aggregation--individual, neighborhood, city, metropolitan area, State, and nation--and the variables used to account for variations in prevalence at each of these levels are apt to be different. For estimating the size of drug user populations, the main data sources discussed are surveys based on questionnaires or interviews and known users' records held by law enforcement, medical, public health, and treatment agencies. Limitations and difficulties associated with using each of these data sources are identified. Estimation procedures, assumptions, and problems associated with these data sources are treated briefly. Log-linear models using multiple samples of known users and procedures for the indirect estimation of prevalence are given particular attention. Examples of prevalence estimates of drug abuse are offered to illustrate different models and procedures. Nine references are listed. (Author abstract modified)