Abstract: Traditional market comparison approach has some problems such as the arbitrariness of trade cases selection and the subjectivity of the weight definition of land price characteristics. A new mechanism of feature weight learning based on Back-Propagation neural network (BPNN) was adopted to solve the above problems. A group of 17-dimensional original features reflecting status of land location and quality were employed as input, and the land prices were used as output of BPNN model. The model could build the complex function relationship fully between the affecting factors and land prices by regulating variable weight connection based on the feedback mechanisms of neural network. The weight coefficient of land price characteristics could be calculated by using the structure parameters of BP network, and has been used in the computing model of similarity degree based on the fuzzy mathematics theory. Then the comparative cases were selected in accordance with the similarity degree. Weighted land price of each selected comparative case was also determined based on the similarity degree. Finally took the Wuhan city in Hubei province as a case study, feasibility and validity of the improvement were validated. The results of the present study indicated that the improved model got better assessment accuracy than the present method with an increase of 3.69 percent.