Aiming at the nonlinear, non-stationary characteristics of a bearing vibration signal, a bearing fault diagnosis method combining improved Intrinsic Time-scale Decomposition based on B-spline(BITD)，Teager energy operator and diagonal slice spectrum was proposed. In this approach, firstly, BITD method was applied to decompose the vibration signal into a finite number of proper rotation(PR) components，then the proper rotation components best representing the fault information were selected to reconstruct signal using the correlation coefficients and the sample entropy, then reconstruct signal was demodulated by energy operator, Finally，the reconstruct signal was analyzed with diagonal slice spectrum to extract the fault characteristics. The method was effectively verified by simulation signals and engineering examples of bearing fault diagnosis.