Housing markets are typically segmented into a number of different sub-markets. If the sub-markets are not included in the hedonic estimation process, parameters will be biased. Furthermore, if neighborhood characteristics are omitted there is a risk that spatial dependency will be present, and this will cause estimates to be biased, inefficient and inconsistent. The objective of this paper is to derive functional sub-markets using cluster analysis to improve upon the hedonic model and reduce spatial dependency. The empirical analysis shows that cluster analysis of the residuals can remedy the problem of spatial autocorrelation. However, if the housing market under investigation is geographically large, the number of clusters will increase rapidly if the objective is to reduce spatial dependency. The predictive performance is highly increased both in the full sample and the testing sample, but the predictive performance will be reduced if the sub-markets created are too small and too numerous. Hence, there is a trade-off between reducing spatial dependency and increasing the predictive power.