A spatial pyramidal cross-correlation based on interrogation area sub-division is introduced to improve the measurement resolution in particle image velocimetry (PIV). The high-resolution velocity can be achieved with a velocity prediction model via coarse cross-correlation. The prediction formula is deduced from the frequency response of the moving average (MA). The performance of this method was assessed using synthetically generated images of sinusoidal shear flow, two-dimensional vortical cellular flow, and homogeneous turbulence. A real PIV experiment of turbulent boundary layer was used to evaluate the new method. The results indicate that the spatial pyramid cross-correlation can robustly increase the spatial resolution.