Urban planners frequently use regression analysis for the empirical estimation of land price models, which are used to investigate the spatial structure of a city or to calculate the implicit price of environmental characteristics. A concern with this kind of analysis is that often spatially associated or heterogeneous data are used, which leads to estimation pitfalls, most notably spatial autocorrelation and model structural instability over space. In this paper, we introduce a modelling methodology that combines a spatial autoregression with a switching regression to incorporate effectively both association and heterogeneity as part of the analysis. In addition, using Sendai City in north-eastern Japan as a case study, we use local statistics of spatial association to assist us in the exploration and definition of heterogeneous regimes. Finally, we show how some measures commonly used to counter spatial effects fail at times, thus highlighting the importance of a spatial modelling methodology such as the one presented here.