The report of the 19th National Congress of the Communist Party of China proposed to deepen the reform of the tax system and improve the local tax system, and the real estate tax is an important part of the tax system. With the deepening of real estate tax reform, how to evaluate the value of real estate has become the focus of real estate practitioners. However, at present, the batch assessment of the tax base of profitable real estate is still limited. The paper makes an empirical study on the office buildings in Shijiazhuang, and demonstrates the superiority of the BP neural network model by comparing with multiple linear regression model.In addition,this paper calculates the expected annual income of the office building and then puts it into the direct capitalization formula to get its taxable value, which provides a theoretical basis for the collection of income property tax.