In response to overcrowded correctional facilities, constrained budgets, and legal pressures, correctional agencies in America are increasingly using offender classification assessment. Such assessments evaluate an offender's risk of future reoffending and/or his/her social and psychological needs. Prominent among recent classification efforts has been the development of the LSI, a promising system of correctional client classification that is of wide interest. The LSI is used most often with community corrections halfway house offenders. A person trained to administer the instrument completes the LSI through a semistructured interview with the inmate. In the halfway house involved in the current study, all those who administered the LSI attended a State training program. Information obtained for the LSI was cross-checked through official records and other sources where possible. The LSI yielded subscores in several areas, including criminal history, education/employment, finance, family/marital issues, accommodation (living arrangements), leisure/recreation, companionship, alcohol/drug problems, emotional/personal issues, and attitude/orientation. Generally, the subscores aided correctional officials in tailoring treatment plans, with higher scores being indicative of greater risk associated with the offender. The current study assessed the predictive utility of the LSI in a population of offenders (n=140) who were directly sentenced to a halfway house in Fort Collins, Colorado. This halfway house has been using the LSI for about 5 years for both client selections, based on perceived risk, and for selecting treatment programs for particular clients. Data were collected on three outcome measures: halfway house outcome; recidivism defined as the occurrence of arrest for any crime for a 2 year period after leaving the residential portion of the halfway house; and recidivism defined as arrest for a felony during the same 2 year follow-up period. Bivariate correlations, Relative Improvement Over Chance (RIOC) analysis, and logistic regression were used to evaluate the LSI's utility. Findings showed that the LSI did not perform as expected in any of the analyses. Possible explanations for these findings are discussed, and future research is suggested. 5 tables, 5 notes, and 29 references