A fuzzy sets-based method for radar weak target classification is proposed and studied. The membership functions of different radar targets are generated by using a two dimensional self-organizing feature mapping network. The kernel estimation of probability distribution and the consistent transformation between probability and possibility are employed by using the network. The problem of feature extraction for radar weak targets is simply discussed. The given testing results show that a high level of robustness and simultaneously a high accuracy of radar ship target classification have been reached by using the proposed fuzzy classification method. The method can be referred in other fuzzy pattern recognition areas.