Targeting the nonlinear and non-Gaussian characteristics of the vibration signals of rolling bearings, an innovative fault analysis approach based on bispectral binary characteristics and FCM method is presented. In this method, firstly, ensemble empirical mode decomposition (EEMD) is utilized to preprocess the signals and decompose the original vibration signal into several intrinsic mode functions (IMFs). Then, the auto-regressive moving average (ARMA) method, which has higher precision and more suitable application conditions, is employed to construct a model for the signal principal components of the IMFs derived from the EEMD. Subsequently, a bispectrum assessment of the ARMA model is carried out. Finally, the binary images extracted from the bispectrum dispersion are regarded as the feature vectors and used to build a classifier of the class templates and the smallest-distance templates through the FCM clustering, thereby achieving the fault analysis. The application results in the fault diagnosis of the outer rolling of bearings confirm that the proposed method is effective as it can accurately determine the actual conditions of the outer rolling of bearings.