Data on defendants used in this study were taken from the D.C. Pretrial Services Agency system. They were collected in connection with an experiment involving periodic pretrial urine testing. All defendants who either had positive urine-test results in the lockup or admitted drug use and who were released on personal recognizance were included in the study. Participants were randomly assigned to one of three pretrial release conditions: periodic urine testing, referral to treatment for drug abuse, or a control group given no special drug-related release conditions. Data analysis involved multivariate statistical techniques ('trivariate probit') specifically designed to address the problem of selection bias. The analysis first controlled for other factors that might effect failure-to-appear, pretrial rearrest, or pretrial misconduct and then considered the additional effect urine-test results would have. There were only a few variables other than urine-test results significantly related to these pretrial outcomes. Analysis by type of drug showed that specific drugs and drug combinations were related in different ways to the risk of pretrial rearrest, failure-to-appear, or overall pretrial misconduct. 8 tables, 4 figures, 37 item bibliography.