In order to implement 3D point cloud scanning of small hole structure, which could not be contacted or damaged, we propose a noncontact 3D measuring method. The system contains a laser triangulation displacement sensor, a Michelson interferometer system and a coordinate measuring machine, with the advantages of non-invasive scanning, fast measurement speed and high precision. Focusing on reconstructing 3D point cloud data, random sample consensus is used to separate surface data and hole data respectively from the raw dataset. Least square optimization determines the function of the cylinder, as well as hole diameter and inclined angle between the hole and the surface. In the experiment scanning a round hole, the estimated result has diameter error and angle error within 30 μm and 0.2°, respectively. Results manifest the effectiveness and feasibility of this system and express practicality in manufacturing industry.