Abstract: Wear properties of six kinds of V9-Cr4-Mo3 high speed steel rolls with different carbon content were tested via cool-rolling wear simulationtesting machine developed by authors.By use of backpropagation (BP) neural network,the non-linear relationship model of the wear weight loss vs.carbon content and wear time was established based on experimental data. The results show that the well-trained BP neural network can predict wear properties of V9Cr4Mo3 high speed steel rolls effectively according to carbon content,and predict wear weight losses precisely according carbon contents and wear time. The prediction results indicate that the matrix microstructure of high speed steel is mainly lath martensite with good toughness and high hardness when carbon content is about 2.58%,resulting in the optimal wear property of roll.If carbon content is too low,the wear failure of roll is mainly caused by severe micro-cutting because the matrix microstructure of high speed steel is ferrite with very low hardness,leading to the decrease of wear resistance.While,the matrix microstructure is mainly composed of plate martensite with poor toughness when carbon content is too high,and therefore it is easy for roll to fatigue,also resulting in poor wear property.