A time-frequency feature extraction method based on center-symmetric local binary pattern (CSLBP) is proposed for vibration acceleration signals of rolling bearings. First, the generalized S transform is employed to process vibration acceleration signals of rolling bearings. Two-dimensional time-frequency images with good resolution performance of the bearing signals are obtained by utilizing the time-frequency aggregation measurement criterion to adaptively set the adjustment parameter of the generalized S transform. Then, the CSLBP of the images is calculated. Texture spectra of CSLBP are extracted and utilized to describe time-frequency characteristics of the vibration acceleration signals of rolling bearings. Vibration acceleration signals from four different rolling bearing states are studied. The experimental results indicate that the texture spectrum of CSLBP can effectively express the time-frequency characteristics of the vibration acceleration signals of rolling bearings. It has low dimensional feature and satisfactory separability compared with the texture spectra of local binary pattern (LBP) and uniform pattern LBP.