The paper presents a novel anisotropic diffusion approach to the problem of ultrasound images denois-ing based on the polar-coordinate representation. Local gradients based on the polar coordinate are introduced and they are more suitable for ultrasound images than horizontal gradients and vertical gradients commonly used in anisotropic diffusion methods. Moreover, an adaptive adjustment scheme for the threshold parameter in con-duction functions is presented according to the incident angle of the ultrasonic beam with respect to the organ surface. Furthermore, based on the structure matrix, an edge-adaptive diffusion model is introduced, which can selectively preserve the edge from the blurring or smooth noise. Experimental results of real ultrasound images show the validity of the presented approach, which takes the physical imaging mechanism of ultrasonic devices into account.