Aiming at the problem that the effectiveness and computation cost of the maximum second-order cyclic stationarity blind deconvolution （CYCBD） algorithm are affected by parameter setting，the crest of the harmonics spectrum （HSC） was used as an evaluation index to determine the length of the CYCBD filter adaptively， and the calculation cost of the optimization process was balanced by the method of encoder time series reconstruction. The bearing fault order is calculated and the cycle frequency is set ac? cording to it. The pulse number of time series reconstruction is determined according to the fault order. The central difference meth? od （CDM） is used to calculate the instantaneous angular speed （IAS） of the reconstructed signal. The filter length of CYCBD was adaptively selected with the HSC as the evaluation index using the equal-step search strategy. The spectrum corresponding to the maximum HSC is calculated to achieve fault feature extraction. The simulation and experimental data analysis results show that the proposed method can adaptively select the filter length， which has an obvious effect on reducing the cost of the CYCBD algorithm， and is effective for rolling bearing fault feature extraction.