In the external evaluation of schools the technique of classroom observation belongs to the methodological standard repertoire. Nevertheless the measurement of quality of classroom teaching based upon selected lesson sequences, which are as a rule inspected only briefly, is fraught with a lot of methodological problems. Therefore it is relevant for a substantiated quality assurance to reveal problems in the measurement of quality of classroom teaching due to an implementation of adequate empirical methods. This is made possible by using the generalizability theory and the many-facet Rasch model. Analyses based upon data of the Hamburg school inspection point out that by using an appropriate data collection procedure rater effects in classroom observations turn out comparatively low at about nine percent of total variance. Furthermore analyses prove that it is insufficient to simply quantify the agreement among raters by using global reliability measures, but that it is necessary to check up on intra rater consistency for getting valid and in this way reliable results from classroom observations for the practice.