Previous optimal thresholding methods are popular and efficient in the case of bi-level thresholding but are computationally expensive when extended to multilevel thresholding to optimise the objective functions. Here, a new method for automatic multilevel threshold selection based on maximum entropy theorem is proposed. The new method has advantages such as greater simplicity, better stabilisation and easier performance compared with some traditional entropy methods, and it calculates thresholds quickly. The approach has been employed to perform image enhancement, segmentation and thresholding, satisfactory results being obtained.