Focusing on the prediction of the wear-in state in wet clutch test, a method based on fractal features of friction torque signals is proposed to predict and evaluate the wear-in state in wet clutch wear-in test. Firstly, calculation methods of box dimension and correlation dimension are introduced respectively. And then friction torque signals are collected from a certain type wet clutch wear-in test, and self-similarity of the signals is analyzed. Finally, the fractal features of friction torque signals are studied through box dimension and correlation dimension. The analytical results show that the fractal features of friction torque signals can be quantitatively described through box dimension and correlation dimension respectively; the fluctuations of box dimension and correlation dimension of friction torque signals decrease gradually with the increase of the wear-in times, and eventually converge to a small range. Therefore the wear-in state of wet clutch can be quantitatively predicted and evaluated by the fractal features of friction torque signals in test under unopened test box condition.