The incipient damageof wind turbine rolling bearingsis very difficult to be detected, because the fault signalsare nonlinear, nonstationary , and likely to be buried by strong background noise. In light of this problem, a comprehensive methodology that combines variational modal decomposition (VMD) and maximum correlated kurtosis deconvolution (MCKD) is presented. The parameters of VMD and MCKD are selected automatically by the particle swarm optimization algorithm (PSO). First, the optimalαand K in VMD are calculated by PSO, and the most sensitive modal is selected according to the VMD decomposition of incipient fault signals. Then, theoptimal L and T in MCKD algorithm are calculated by PSO so as to boost the fault shock in the modal. Finally,the incipient fault feature is extracted from the envelope demodulation of the faults. Simulation results as well as experimental tests both validate that the proposed method can adaptively enhance the weak fault component of rolling bearing, thus can effectively extract incipient fault features of rolling bearings from strong background noise.