The UK health care reforms, leading to the internal National Health Service market, have created pressures to assess the performance of hospitals whether in terms of the efficiency with which they use their resources or the outcomes they achieve. There is a tendency to assume that ‘least is best’, but it should be fairly obvious that the performance or outcomes of a specialty in a particular hospital will depend on the characteristics of its patients, its diagnostic case mix and the social circumstances of the areas in which its patients live. Summary measures which do not take account of these influences are inevitably suspect. Recent controversy over the UK Government's ‘star ratings’ mirrors similar arguments over the publication of league tables for schools; this paper uses the statistical method of multilevel modelling (developed in the context of assessing educational performance) to explore variations in the length of hospital stay for gynaecology patients in Scotland. The conclusions emphasize the importance of distinguishing between influences that can be reasonably attributed to the hospitals themselves and those which arise from differences in the patients they treat or the areas from which these patients come.