On the basis of characteristic of differential “curl to straight” and disperse calculation method, an extraction method using local reconstruction and differential calculation was proposed for valley-ridge features from point clouds. First, the dispersed Laplacian operator was used to enhance the point data and an appropriate threshold was set to obtain potential valley-ridge points. Then, a triangle mesh attached on the potential surface was constructed in every valley-ridge points to effectively reflect the local geometry feature information. Finally, according to Weingarten mapping, principal curvatures and principal directions of potential valley-ridge points were calculated. A simple differential calculation method and a linear interpolation method were used to decide if the point is an extreme value point in principle direction and to extract the valley-ridge features based on the multi-scaling idea. The experimental results indicate that when the number of vertexes and the number of valley-ridge features are 10 375 and 1 129,respectively, the execute time is just 97.39 ms. And when those are 327 853 and 105 482, the execute time is 3 956.12 ms. The method proposed in this paper is simply, stable, and avoids higher time cost due to fitting surface to approximate the differential quantities in the traditional method, so it extracts valley-ridge features from point clouds fast and efficiently.