Accounting for spatial interactions between farms is highly relevant for the analysis of agricultural policy impacts. Existing studies, however, only account for homogeneous (average) spatial interactions. We develop a mixture modelling framework to account for unobserved heterogeneity, allowing spatial interaction to vary across endogenously-defined farm types. We apply this approach to study farmer decisions to exit the farming business using a large panel dataset covering all registered farms in Brittany, France, for the period 2004-2014. While exiting is on average positively correlated with neighbouring farm size, we find substantial variation between farm types, and a negative correlation for a significant proportion of farms. The approach we develop not only enables us to identify different correlations between neighbouring farm size and exits from farming, but it also yields different results than pooled estimations.