Forest inventory automation has become a major issue in forestry. The complexity of the segmentation of 3D point cloud is due to mutual occlusion between trees, other vegetation, or branches. That is why, the applications done until now are limited to the estimation of the DBH (Diameter at Breast Height), the tree height and density estimation. Furthermore other parameters could also be detected, such as volume or species of trees (Reulke and Haala) . . . This paper presents an effective approach for automatic detection, isolation of trees and DBH estimation. Tree isolation is achieved using an innovative approach based on a clustering methodology followed by a skeletonization step. The DBH of trees is then determined automatically. The efficiency of our algorithm is evaluated with comparison with ground data, measured by classical methods.