Abstract: Objective An improved water-fat separation method based on region –growing was proposed for use in regions with low signal-noise ratio (SNR). Methods Region-growing method was applied to 4 sub-images acquired by a downsampling operation on the acquired phasor maps. The spatial smoothing constraint was exploited to calculate 4 error phasor maps to construct the final smooth error phasor map, which was used in two-point Dixon technique for water-fat separation. Results The simulation experiment showed that the proposed method produced smaller errors, and for clinical images of the knees, abdomen and lower limbs, the proposed method achieved accurate water-fat separations. Conclusion The proposed method is more robust and reliable than the original global region-growing algorithm, and serves as a promising water-fat separation method for clinical applications.