The present study uses the Mixed-Rasch Model to analyze questionnaire data. It is hypothesized that more than one latent sub-population (classes) can be identified. It is also hypothesized that response latencies and social desirability can be used to describe and discriminate the latent classes. Using the software WinMira 2001, data from the Eysenck-Personality-Profiler-German-version (EPP-D) was analyzed. A so-called non-scaleable and a scaleable group of subjects were identified which differed with regard to the use of one response category, I don't know. However, there was no difference between these groups with regard to response latencies and social desirability. Results will be discussed with regard to the EPP-D scoring scheme as suggested in the manual.