By introducing grayscale preference parameters and gradient valley coefficients, an improved grayscale preference valley emphasis (GPVE) method is proposed. The grayscale preference parameter makes the threshold show a tendency to approach the object gradation, and the gradient valley coefficient makes the threshold closer to the valley. GPVE method can achieve an ideal threshold when the histogram is unimodal and bimodal. The experimental results show that the average segmentation error ME value obtained by Otsu, VE, NVE, WOV and other methods is above 0.2, while the ME value of the proposed method is only 0.007. Thus the proposed GPVE method improves the accuracy of image segmentation and realizes a better segmentation effect.