The research explored two general mechanisms, including investigator-initiated termination and subject-initiated withdrawal. Subjects were terminated when patterns of incoherent or inconsistent responses were detected in interviews; subjects were also terminated when they stole from the field site or when they were jailed or hospitalized. The findings suggest that the initial phases of interviewing served as an important subject screening period for the researchers who conducted the interviews. Most noncompletion in both studies fell under the category of subject-initiated withdrawal. Analyses suggested that respondents who dropped out reported significantly less drug involvement and somewhat less extensive criminal involvement than those who completed the study. Women who dropped out were more likely to be married than women who completed the study. Men who dropped out were less likely to reside in shelters for the homeless than men who remained in the study. The behavioral findings sharply contrasted with findings from studies of attrition in conventional panel research on drug use. Implications for research on hidden populations are discussed. These findings suggest some important limitations in ethnographic research and underscore the importance of systematic descriptive accounts of noncompletion across a range of ethnographic studies. 4 tables, 5 notes, and 29 references