An investigation was made of the error-choice principle as a possible alternative to the usual form of direct questioning used in personality questionnaires. A standard personality questionnaire consisting of 91 items of the ‘yes— ?—no’ type was administered to a sample of 150 subjects (normals, neurotics and psychosomatic patients). The questionnaire, which comprised scales for neuroticism, extraversion, test-taking attitudes, masculinity and social desirability, was administered along with an error-choice version of the same items. The latter consisted of items followed by two alternative percentages (‘representing’ the p-values in the general population) each of which was equally wrong, but in opposite directions from the correct p-value. The subject had to circle the percentage that he believed to be correct. The error-choice inventory was presented to the subjects under the disguise of being a test of subject's ability to be insightful of people. Of all 91 items, 90 showed positive correlation between endorsement in the self-report format and over-estimation in the error-choice version. These correlations were significant at the 5 per cent level in 73 of the 90 items. A further analysis of items belonging to the distinct scales and of the various response categories was carried out and the operation of some response sets was investigated.