The noises, which come from the 3D Point Cloud data obtained by the Heidelberg Retina Tomograph (HRT), can be smoothed effectively by a bilateral filtering algorithm. This algorithm can retain the graph feature information while denoising, but the execution time of this algorithm increases greatly with the increase of the iteration number, so that this algorithm can not be applied to the diagnostic  practice. The mean neighborhood method can also smooth the graph through the average value processing to a point's Z coordinate of a certain neighborhood, and selecting different weights according to the distance from the target point, but the effect is not better than using the  bilateral filtering algorithm. So this paper presents the mean neighborhood method to perform preprocessing by bilateral filtering denoising. We find our method can significantly reduce the computing time.