Analysis and pattern recognition of the interference undergoing in the communication system can assist the self-adaptive adjustment of the communication system parameters, thereby the anti-jamming capability can be stronger and targeted. A wide-bandwidth communication system is researched. Previous research shows that multi-hidden-layer neural network can resolve any form of classification problems. In order to classify the five common interference patterns, a classification method which uses power spectrum density and two-hidden-layer neural networks is proposed. Simulation results show that, under different interference patterns and different Interference-Noise-Ratios(INR), the average recognition accuracy is above 99.6%. In all the other four interference patterns without comb-spectrum interference, the recognition accuracy is above 99.7%, while 98.4% in the comb-spectrum interference. The proposed method has relatively stable recognition ability, and can be applied to the detection of communication interference.