When the image histogram is a single peak, the traditional Otsu method can easily obtain wrong results in detecting defect in the electrowetting display. In some improved methods, the segmentation results are unstable when the defect color depth is different, and the contrast between the background and defect is low. In this study, an improved maximum between-class variance method is proposed to solve these problems. To improve the difference between the peak and non-peak ranges of the histogram and to better extract the peak information, the proposed method adds a weight value that decreases with an increase in the cumulant of gray histogram gradient before the target variance. It ensures that the threshold obtained by the method is always on the left side of the single peak for a single peak. Experimental results showed that the average misclassification error of the proposed method in multiple application scenarios is reduced by 0.4781 compared with the traditional Otsu method. Besides, the average defect segmentation rate of the proposed method is increased by 0.6795. The method can successfully segment electrowetting display defect and various types of defect. The segmentation effect is better when the contrast between the defect and background is low.