As health services research focuses on prevention and drug treatment programs for special populations, maintaining panel samples at adequate levels over time becomes more important. Differential panel mortality poses a serious threat to the internal validity of experimental designs and the external validity of study conclusions. Minimizing respondent attrition demands systematic attention to two problems: location of subjects at each wave and sustaining respondent cooperation over the life of the research. Major obstacles to collecting longitudinal data on non-traditional populations include the inapplicability of most community-based and governmental records, inexperience with tracking those without stable addresses, and cost. This paper describes and evaluates methods used to track a panel of 670 homeless substance abusers over 3, 6, and 12 month intervals. Results suggest that a broad brush tracking strategy with emphases on telephone, mail and field tracking served to minimize panel mortality and its accompanying methodological problems.