It has been widely agreed in real estate literature that housing submarkets should be adopted as a working hypothesis. This paper revisits the housing sub-market hypothesis and seeks to spatially delineate residential market segments using Hong Kong geo-coded transaction data. Spatial correlation information inherent in the Hedonic residuals is adopted to define property clusters, with which we define sub-market segments. Chow test is performed across the clusters to examine the segmentation of the market. The results are then compared with the prevailing structure of market segments. Weighted mean square test reveals that the defined sub-market structure can improve the accuracy of price forecast significantly, which is consistent with previous research findings.