Aiming at the problem of fault diagnosis in diesel engine valve trains, a fault diagnosis method was proposed based on Wigner distribution and differential box-counting fractal dimension. First, the improved local mean decomposition (LMD) was used to decompose the vibration signals of the cylinder head into several product function (PF) components, and the correlation analysis was selected to eliminate noise and pseudo components. Second, for each relevant component, Wigner distribution was calculated separately and then accumulated to construct the time-frequency image of the vibration signals. Then, the differential box-counting fractal dimension was extracted as the fault feature. Finally, the k-newerest neighbor algorithm (k-NN) was used to fulfill the fault diagnosis task of diesel valve trains. The simulation results showed that the improved LMD method efficiently suppressed the cross-term of Wigner distribution. The experimental results showed that the differential box-counting fractal dimension was superior to the other six kinds of typical fault characteristics, and the fault diagnosis accuracy was 97.2%. Therefore, the proposed method can be used to diagnosis the fault of diesel engine valve trains.