Clustering individuals or households on the basis of socio-economic variables has become a widespread practice in German social research over the past few decades. This paper is part of a research project that explores results which may be obtained when time use patterns are chosen as the basis of numerical classification. Over the past few years, results relating to single households were published by the authors. The present paper extends the analysis to families. The investigation uses data from the German Time Use Survey 2001/2002. It is shown that the clustering process fulfils the criteria required by stochastic and qualitative social science. Furthermore, evidence is provided that including cluster memberships as dummy variables into a regressor set increases the predictive capabilities of a common multivariate analysis of correlations between socio-economic variables. Especially concerning health, meaningful interconnections between household styles and health state are detected.