Levels of obesity and overweight in the UK are high with certain groups within the population particularly affected. The customary approach is to identify at risk groups based on their socio-demographics or their observed unhealthy food choices. This approach fails to acknowledge that households with similar socio-economic background may behave very differently and that households make unhealthy food choices for very different reasons. In this study we segment households according to their underlying food preference using an Almost Ideal Demand System (AIDS) mixture model. We identify five household segments that are similar in their food preferences and therefore in how they would respond to policy interventions. The food purchasing patterns of households in each of the five segments tend to be similar but households differ in terms of socio-demographics. This information needs to be taken into account when designing a targeting mechanism for policy interventions to improve diets.