The method uses maximum likelihood estimation techniques. Based on suggestions by Lee (1984), a multivariate probit estimation technique was implemented to allow estimation using data from the Washington Pretrial Services Agency for 2,311 arrest cases; 1,824 of these were released either on bail or recognizance, with or without conditions. Results of the analysis demonstrates the feasibility of the probit technique. A test of selectivity bias was conducted by estimating a variety of models using conventional, single-equation techniques and comparing them to the multivariate probit results. This comparison suggests that selection bias is substantial in the conditional estimates of behavior in the pretrial justice system. For example, conditional estimates give the impression that release on bond has little or possibly a negative effect on failure to appear. The unconditional estimates, conversely, suggest that release on bail does act as a deterrent of failure to appear, but not of pretrial arrest. Decisions on pretrial release classification criteria, particularly on the overall level of expected misconduct, therefore should be made using unconditional estimates. The multivariate probit techniques developed here can provide such estimates for classification and policy development purposes. Supplemental data are appended, including variable names and descriptions. Tables, figures, and 33 references.