A multitrait-multimethod design was used to examine the convergent and discriminant validity of seven pain measures from three widely used self-report instruments designed to assess the sensory, affective and intensity dimensions of pain. The instruments were the McGill Pain Questionnaire, the Pain Perception Profile and Numerical Ratings. Three distinct factor models, each corresponding to a different hypothesis about how these pain measures are related, were tested using confirmatory factor analysis in a sample of 419 headache sufferers. A three-factor model, postulating three correlated factors defined by the three assessment instruments best explained the correlations between the pain measures. Measures of sensory, affective and intensity dimensions from the three instruments failed to exhibit convergent or discriminant validity. Rather, instrument variance obscured the pain qualities the three pain instruments were designed to assess. These findings suggest that greater attention needs to be paid to how formal characteristics of pain assessment instruments influence patients' descriptions of their pain.