Measures of personality disorder from the International Personality Disorder Examination, Personality Diagnostic Questionnaire, and Millon Clinical Multiaxial Inventory (MCMI-II) were obtained from detained male mentally disordered offenders (N = 156), and convergent and discriminant validity were examined by confirmatory factor analysis of the multitrait-multimethod matrix. Hierarchical comparisons of models varying in their specification of trait and method variance established the appropriateness of a model supporting both convergence and discrimination across methods, but these were variable across constructs and measures. Convergence was good for avoidant, schizoid, and antisocial disorders, but poor for histrionic, narcissistic, and obsessive-compulsive disorders. Avoidant, schizoid, and schizotypal disorders were not clearly distinguishable from each other. Measurement error attributable to method variance was substantial for all instruments and for most disorders. The commonly alleged superiority of interview over questionnaire methods was not supported, and the MCMI-II demonstrated proportionately more "true" variance. However, assessment methods may be differentially sensitive to different kinds of personality disorder problems.