The Level of Service Inventory-Revised (LSI-R) assesses the risk of reoffending and identifies criminogenic need characteristics of offenders. Despite the assessment's popularity, its factor structure remains under-researched. Inconsistency in published research may be attributed to the misuse, and inappropriate statistical analyses, of the LSI-R subscale scores. The present study explored the factor structure of the LSI-R, at the item level, using a sample of Australian offenders. Binary exploratory factor analysis found a five-factor solution for male offenders and a four-factor solution for female offenders. Using these data to "recalibrate" the LSI-R, sensitivity and specificity comparisons of the original and modified LSI-R were undertaken. With specific attention to Indigenous status and types of sentence orders served, sensitivity and specificity improvements were observed. Findings are discussed in terms of the value of exploring latent constructs and validating assessment tools, as well as the potential impact on the management and rehabilitation of Australian offenders.