It is difficult to identify the fault state and location of the roller bearings of a wind turbine due to the complexity of its vibration signals. They suffer strong interference and are non-Gaussian, nonlinear and non-stationary. In the light of these characteristics, a fault diagnosis method is put forward based on the binary-bispectrum and fuzzy clustering. First, the binary-bispectrum features are obtained. Then, the target templates for different faults are constructed based on fuzzy clustering method. Finally, the proximity classifier is designed to determine the roller bearing fault position according to the distance between the test samples and corresponding target templates. The experimental results show that this method can effectively diagnose the fault state and fault location with high accuracy, good stability, less calculation and fast speed. Furthermore, the results is easy to understand and verify when the relative distance is taken as the basis to diagnose the faults.