The nonlinear GRAPPA method was applied to estimate the undersampled K-space data in order to overcome the inaccuracy of sensitivity map of each coil calculated by zero-padding method. Then the accurate sensitivity map of each coil was calculated. The sensitivity encoding (SENSE) method was adopted to reconstruct the MR image from the under-sampled K-space data based on the accurate sensitivity map. The proposed hybrid technique was tested on MR brain image reconstructions at various acceleration rates. Experimental results show that the proposed method can achieve significant improvements in reconstruction accuracy when compared with the nonlinear GRAPPA and SENSE methods, with lower artificial power (AP) and higher signal to noise ratio (SNR).