Comparative quality analyses of rehabilitation centres are required by the legislators and are the focus of the external Quality Assurance plans presently implemented. However, they are also highly relevant for internal Quality Management models (e.g. for the result criteria of the EFQM model). To control for confounders of rehabilitation success that cannot be influenced by the rehabilitation centre (e.g. age, co-morbidity), and thus to permit fair comparisons of clinics, regression analysis risk adjustment procedures are primarily used in the literature. The present paper explains the use of so-called Hierarchical Linear Models (HLMs) using example of data of N = 2.044 patients undergoing rehabilitation following hip and knee operations from the Quality Assurance programme of the statutory health insurance funds (QA-Reha-procedure). This procedure has the advantages of: a) taking into account the multi-level structure of the comparison problem; b) permitting the inclusion of predictors at the rehabilitation centre level; and c) permitting the modelling of variation in regression coefficients over the centres. The data presented show that the differences in achieved rehabilitation outcome among the rehabilitation centres - after control of the confounders by means of HLMs tend to be slight. In addition to patient-related predictors of rehabilitation outcome (baseline somatic, functional, psychosocial status, co-morbidity, rehabilitation motivation, gender, age), the mean functional disability of the patients in the centre is shown to be a confounder at the clinic level. In this respect, a centre that has little experience with severely affected rehabilitation patients achieves on average lesser effects on somatic, functional, and psychosocial levels.