A method based on point cloud smoothing approaches for detecting noise and outliers is introduced. This method firstly estimates thresholds according to points’ shifts after smoothing, secondly identifies outliers and noise whose shifts are more than the thresholds and lastly removes them and repeats the whole process. The main difference from other methods is that it tries to screen out outliers from point clouds rather than makes points smoother by correcting their coordinates. The precondition to assure it work is that outliers will be shifted more, which requires smoothing operators possess the ability of low-pass filter. Since this method does not rely on any local parametric representation, it can deal with noisy data in the case of ambiguous orientation and complex geometry. The proposed method is rather simple and can be easily implemented. Some cases are provided to illustrate its feasibility and effectiveness.