In order to exclude the effect of dissimilar genes, a new missing value estimation method based on KNN-SVR is proposed. This method selects a group of complete genes most similar to target genes by K-nearest neighbor (KNN) and uses them to estimate missing values by Support Vector Regression (SVR). This paper also suggests using the variance of Normalized Root Mean Squared Error (NRMSE) to measure the stability of estimation methods and the reliability of estimated values. This method improves the validity of missing value estimation by filtering genes. The experiment results show that KNN-SVR method has better accuracy and stability.