In order to improve the convergence performance of the normalized subband adaptive filter (NSAF), the variable step-size affine projection normalized subband adaptive filter (VSS-AP-NSAF) based on multi-band structure is proposed in this paper. The VSS-AP-NSAF updates the tap-weight vector based on several previous input vectors, which not only increases the convergence rate, but also maintains the robustness against the impulsive interference. The variable step-size is obtained by minimizing the upper bound of mean square deviation in iteration. In addition, the VSS-AP-NSAF is similar to the affine projection algorithm (APA) on the computational complexity. Experimental results demonstrate that the VSS-AP-NSAF achieves faster convergence rate and lower misalignment than the existing algorithms in system identification, and has a good result for echo cancellation as well.