A laser point clouds data association decision algorithm is proposed. The multi-geometric feature extraction and intelligent management for the laser point clouds are investigated with the discriminative graphical model. The maximum pseudo-likelihood learning is employed to optimize the weights of the local and pairwise features. And the states of the hidden nodes in the graph are estimated with max-sum probabilistic inference. Furthermore, the laser point association is tackled as the maximum a posteriori (MAP) configuration backtracking problem. The experiment results demonstrate that the proposed algorithm outperforms traditional algorithms.