Background Comorbidity research in psychiatric epidemiology mostly uses measures of association like odds or risk ratios to express how strongly disorders are linked. In contrast, chronic disease epidemiologists increasingly use measures of clustering, like multimorbidity (cluster) coefficients, to study comorbidity. This article compares measures of association and clustering. Methods Narrative review, algebraical examples, a secondary analysis of an existing dataset and a pooled analysis of published data. Results Odds and risk ratios, but the former more than the latter, confound clustering with coincidental comorbidity. Multimorbidity coefficients provide a pure estimate of clustering which is the proportion of the association between disorders that is of etiological interest. Odds and risk ratios can express comorbidity between no more than two disorders, whilst clustering coefficients, although computationally laboursome, can capture multimorbidity of any number of disorders. Cluster coefficients depend less on the prevalence of illness in study groups than measures of association. Conclusion Odds and risk ratios are well suited for comorbidity research which focuses on which sets of disorders or syndromes tend to occur in combination and the implications of this for, for instance, nosological classification, a traditional interest of psychiatric epidemiology. However, the cluster coefficient is to be preferred if the interest is more aetiological, addressing for example why certain individuals are prone to multiple health problems.