Abstract: A new SNR estimation algorithm based on relevance vector machine (RVM) was proposed, it can not only be used under the flat fading channel, but also meet the request of larger estimation range and higher estimation accuracy of PSK. The estimation model was created by using RVM, based on the relation between the SNR and the two-order moments and four-order moments, and the reliable weights of the model can be figured out by straining and studying. Simulation results show that this algorithm has lots of advantages, for example, it uses less data and has larger estimation range than general algorithm, in addition , its estimation accuracy becomes higher in the effective estimation rang, and it is applicable to more modulation signals.