This paper discusses, estimates and formally compares the best known procedures for incorporating demographic variables into complete demand systems. In particular, a class of general procedures belonging to Gorman's family of ‘general linear household technologies’ is introduced. Estimation and comparison of different procedures make use of Italian household budget data for the years 1973–1992, incorporating a single demographic variable (family size) into a Generalized Quadratic Almost Ideal Demand System. In our empirical example, however, even the most general household technologies are unable to fully capture the behavioural heterogeneity shown by the data. Copyright © 2000 John Wiley & Sons, Ltd.