Two self-report measures of psychopathy, Levenson's Primary and Secondary Psychopathy scales (LPSP) and the Psychopathic Personality Inventory (PPI), were administered to a large sample of 1,603 offenders. The most widely researched measure of criminal psychopathy, the Hare Psychopathy Checklist-Revised (PCL-R), served as a provisional referent for estimating the construct validity of these self-report measures with offenders. Compared with the LPSP, the PPI displayed higher zero-order correlations with the PCL-R, better convergent and discriminant validity, and more consistent incremental utility in predicting PCL-R scores. Furthermore, using a variant of Westen and Rosenthal's approach to evaluating the construct validity of a new measure, compared with the LPSP, the PPI's pattern of associations with measures of 35 external criterion variables was more similar to the pattern observed for the PCL-R. Results generally provide stronger support for the validity of the PPI than the LPSP in offender populations using the PCL-R as a provisional benchmark, particularly for assessing interpersonal and affective features of psychopathy. (Contains 5 notes, 2 tables, and 1 figure.)