The study tested three structural models for goodness-of-fit of data obtained from a sample of subjects (n=206) receiving drug misuser treatment in four treatment settings in a treatment program in Nevada. Fourteen variables were entered into the structural equations. An adequately fitting model was found. The subjects' discharge rates did not differ significantly by the severity of cocaine misuse at admission. The study showed how latent variable structural equation covariance models support a new paradigm for cocaine misuser treatment outcome research, but their utility in explaining the variance of treatment implementation by pretreatment status variables needs further investigation. Limitations of the study include missing cases for repeated measures and non-normally distributed data. While generalizability of findings from normally distributed populations is a tenable assumption in multivariate theory, the observed data from this sample were not generalizable to all cocaine misusers. A lack of data on treatment process left unresolved issues about the interactions between cocaine misusers and their therapists. Tables, figures, references