We maintain that the appropriate definition of submarkets depends on the use to which they will be put. For mass appraisal purposes, submarkets should be defined so that the accuracy of hedonic predictions will be optimized. Thus we test whether out-of-sample hedonic value predictions can be improved when a large urban housing market is divided into submarkets and we explore the effects of alternative definitions of submarkets on the accuracy of predictions. We compare a set of submarkets based on small geographical areas defined by real estate appraisers with a set of statistically generated submarkets consisting of dwellings that are similar but not necessarily contiguous. The empirical analysis uses a transactions database from Auckland, New Zealand. Price predictions are found to be most accurate when based on the housing market segmentation used by appraisers. We conclude that housing submarkets matter, and location plays the major role in explaining why they matter.