Abstract: Machine learning, as a learning method, uses the experience to improve its performance. The support vector machine (SVM), as a new model of the machine learning, is good at dealing with the condition of small sample size, nonlinear and high dimensional pattern recognition. The node localization algorithm based on SVM can locate the nodes in wireless sensor networks, WSN depending on the characteristics of machine learning algorithms. The basic idea is dividing the network area into several small aliquots of grids and each represents a certain class of machine learning algorithm. And when the machine learning algorithm learns the classes corresponding to the known beacon nodes, it will classify the unknown nodes’ localization and then further determine the position coordinates of the unknown nodes. For the SVM “one against one” location algorithm, the simulation results show that it has higher location accuracy and better tolerance of the ranging error, which is suitable for the network environment where the beacon nodes are sparse as it doesn’t require a high beacon node ratio. For the SVM decision tree location algorithm, the results show that it is not affected seriously by the coverage holes, which is applicable for the network environment where nodes distribution is uneven or the coverage holes exist.