Real estate economists and practitioners have long been aware of the problems arising from spatial autocorrelation in housing prices in undertaking housing appraisal. To tackle them, various estimation methods and techniques such as spatial lag model and spatial error model are devised and tested empirically in order to enhance appraisal accuracy and consistency. Interestingly, the underlying factors affecting the formation of spatial autocorrelation have so far received little scholarly attention In this paper, we put forward a novel hypothesis that spatial autocorrelation in housing prices is a result of the price information search process undertaken by house traders: Spatial autocorrelation emerges when they infer prices of one housing unit from another; Moreover, we conjecture that building age, which is an indicator of the buildingís re-development potential, serves as a factor attenuating spatial autocorrelation since traders rely less on price information of aged housing units, which are prone to be rebuilt, to establish current market values of other units. In other words, it is believed, and hypothesized, that spatial autocorrelation decreases as building age increases. We test the above hypothesis using geo-coded open market transaction data of Hong Kong. The results not only confirm the role building age played in the price discovery process in real estate, but also in the formation of spatial autocorrelation of housing prices.