The use of scales on the Personality Assessment Inventory (PAI) to detect defensiveness in criminal and nonclinical samples was evaluated. Forty-five male inmates of a county jail and 38 male undergraduate psychology students were provided with incentives to complete the PAI under two conditions: standard instructions and experimental instructions to feign a specific, positive role. The sequence of instructions was counterbalanced in both samples for the purpose of examining ordering effects. A repeated measures multivariate analysis of variance (MANOVA) was performed, yielding significant main effects of condition, group and order. Additionally, a step-wise discriminant function analysis significantly predicted group membership (i.e., subjects under honest and faking conditions) with a hit rate = 84.4%. Finally, a more effective cutting score for the Positive Impression scale was recommended.