To overcome the shortcomings of traditional methods, a method of underwater image segmentation based on the discrete fractional Brownian random field was proposed to dispose underwater images. At first, a window was set up, and the centre of window was located at the position of each pixel in the image. The average of fractal dimension in the window was calculated, and it was considered as the fractal feature of the pixel at the centre of window. At last, a threshold was determined according to the graph of fractal dimension, and the segmentation of underwater image was completed. By the normalization of the average absolute intensity difference on surfaces at difference scales, the number of data items used to represent the average absolute intensity difference on surfaces at difference scales was reduced, and the segmentation efficiency was improved. Finally, the results on some typical images were presented. Compared with the results obtained by the segmentation methods of Ostu and Maximum Entropy, it shows that the presented method is robust and efficient in underwater image segmentation.