A novel method was proposed to estimate the normal for a scattered point cloud with sharp features to overcome the shortcomings that existing methods are unable to reliably estimate the normal of point cloud model and lead to the smoothed sharp features. With proposed method, the normal of point cloud was estimated with principal component analysis method. Then, different values were weighted on neighborhood normals according to spatial distance and normal distance of current points of the neighborhood, and the revised or current normals were updated by the sum of weighted neighborhood normal. Finally, the average deviation between standard normal and estimated normal was measured and the accuracy of estimated normal was evaluated. The estimated normal was applied to point cloud processing to verify the feature-preserving property. The proposed method was validated. The results demonstrate that proposed method accurately estimates the normal for data with noise and the least average deviation is close to 0. Moreover, the method has good robustness to the niose, and it keeps the original geometry well when the normal is used as input of the point cloud processing. Comparing with other sharp feature preserving normal estimation methods, the proposed method shows smaller average deviation, higher processing speeds and less computation time.