In order to extract the characteristics of frictional vibrations and realize the recognition of friction states for friction pair,the friction and wear tests of mixed friction and dry friction states of the friction pair were carried out using a reciprocating friction and wear testing machine.Frictional vibration signals collected during the tests were denoised by applying the spectral subtraction,and 15 characteristic parameters of the friction vibration after noise reduction were calculated.The neuron classification of the self-organizing map (SOM) neural network was obtained by analyzing the characteristic parameters of the frictional vibration of friction pair in different friction states with SOM neural networks.The results demonstrate that the spectral subtraction method can eliminate the background noises of the friction and wear testing machine,and the SOM neural network algorithm can effectively analyze the features of the frictional vibration signals and realize the recognition of the friction state of the friction pair.