Excess weight is a problem affecting over half of the British population, with some categories being more at risk than others, in particular in lower socio-economic groups. In that respect, differentiated dietary behaviours are known to contribute to inequalities in health outcomes. Segmentation is increasingly employed as a means of better targeting policy interventions. While conventional segmentation methods divide the population according to their dietary choices or according to socio-demographic characteristics, a potential flaw in this approach is that people may choose to consume a bad diet for entirely different reasons, or that people from different socio-demographic groups may behave in a similar fashion. We use a novel alternative approach which seeks to segment according to peoples dietary preferences. The method estimates a finite mixture of AIDS. We identify segments which have a degree of homogeneity in their food purchases, while remaining heterogeneous in terms of their socio-demographics. The homogeneity of food purchases within components is less than within components identified using k-means clustering of food choices. We argue that this approach will lead to more effective targeted interventions because they would appeal to the reasons for bad dietary choices rather than the choices themselves.