In view of the problem that the existing real estate valuation models do not include spatial autocorrelation and nonlinear influencing factors, a partially linear spatial autoregressive model that can flexibly explain the meaning of variables is proposed to fit real estate valuation data. For the estimation of partial linear spatial autoregressive models, the two-step estimation process combining the local polynomial method and the quasi-maximum likelihood estimation method is used to obtain the estimation of the parameter part. The fitting result of real estate valuation data shows that real estate valuation data does have spatial correlation, and the distance between the house and the nearest MRT station is negatively correlated with the housing price, while there is a positive correlation between the number of convenience stores in the living circle on foot and the housing price, which is similar to the practical explanation. In addition, the nonlinear relationship between the age of house and the housing price is also reflected. Partially linear spatial autoregressive model can explain the practical significance of real estate valuation data more objectively and flexibly.