Choice experiments have become a widespread approach to non-market environmental valuation. Given the vast range of public opinions towards environmental management changes, it is desirable that analysis of discrete choice data accounts for the possibility of unobserved heterogeneity amongst the population. There is, however, no consensus about the best way to model individual heterogeneity. This paper presents four approaches to modelling heterogeneity that are increasingly used in the literature. Latent class, mixed logit, scaled multinomial logit and generalised mixed logit (GMXL) models are estimated using case study data for catchment environmental management in Australia. A GMXL model that accounts for preference and scale heterogeneity performs best. I evaluate the impacts of models on welfare estimates and discuss the merits of each modelling approach.