OBJECTIVES: The Positive And Negative Syndrome Scale (PANSS) and Calgary Depression Scale for Schizophrenics (CDSS) are widely used in the evaluation of schizophrenia. Their internal validity have already been evaluated with classical methods (multitrait and confirmatory analyses). During the last two decades, Item Response Theory has been developed to deal with latent traits. As part of it, Rasch models are commonly used in Quality of Life research but not yet for other outcome questionnaires. METHODS: 458 schizophrenic patients were evaluated with the PANSS and CDSS. Rasch models for polytomous items were fitted to the data in order to assess: 1) unidimensionality of the CDSS and the PANSS subscales, i.e. their ability to measure one latent trait (degree of depression / degree of positive, negative, general schizophrenic symptoms); infit and outfit statistics were used and residuals studied; 2) invariance of comparisons, implying that the parameter characterizing an item does not depend on the latent trait distribution of the population; item parameters estimates were compared for two subgroups of the population. RESULTS: Unidimensionality and invariance of comparisons are globally satisfactory for the CDSS, although the appropriateness of two items (items four and seven) may be questionable. Results do not support the three-dimensional structure for the PANSS, which is commonly used as the reference. CONCLUSIONS: Further investigation of the factorial structure of PANSS (e.g five-factors structures, which have been proposed by several authors) is necessary. Rasch models provide a powerful approach to evaluate internal validity of mental health scales, enabling to investigate invariance of comparisons, which constitutes the major distinction from classical methods.