Land value sometimes needs to be extracted from the price of a developed property for a number of purposes. For instance, land value tax has long gained recognition of being an efficient and equitable tax. Despite the theoretical significance, daunting difficulties in valuing land often cause land value tax practically impossible. The extraction of land value is however required by legislation in countries such as Taiwan, Japan, and Australia, among others. A number of recent studies have explored the feasibility of constructing empirical models that particularly tackle this issue. This article, using a data set of residential property sales over the period of 1994 to 2003 in Taipei City, Taiwan, provides another attempt. In order to alleviate the difficulties of estimating the value effects of spatially-associated property features through a series of independent variables (distances to service facilities, neighborhood amenity, etc.), a trend surface app roach is employed to establish a locational or spatial variable. This locational or spatial variable aims to capture the overall effects of spatially associated variables on property value. The percentage of land value in the total property price is derived from both linear and non-linear regression models. Results are compared to the rule-of-thumb figures currently adopted by the appraisal profession. The differences, if found, in the percentage of land value as the property price between research findings and the rule-of-thumb are expected to have policy significance.