This paper describes a fast reconstruction algorithm of implicit model based on 3-color octree structure for dense unorganized point cloud. At first, the point cloud is stored with an extended octree, 3-color octree. Aiming at this 3-color octree structure a new node watershed algorithm is presented with a higher efficiency to estimate the signs of subdivided leaf nodes. So the leaf nodes are divided into three types: interior, boundary and exterior nodes. To quickly reconstruct the model we sample the 3-color octree structure only at boundary nodes, which greatly reduces the number of sampled points. Then, the triangular meshes are extracted according to the relationships of boundary node. Finally the applications are illustrated in several point clouds, which shows the efficiency and precision of this reconstruction algorithm.