The ultrasound image is significantly degrade by the coherence speckle. A pyramid based multiscale anisotropic diffusion filter is proposed for medical B-mode ultrasound image despeckling. In the Laplacian pyramid domain, according to the different bandpass ultrasound image characteristics in each layer, the anisotropic diffusion filter is adaptively performed to suppress the speckle. For the anisotropic diffusion, the speckle scale function is estimated in the homogeneous area in each layer and the number of iteration is automatically determined by the Mean Absolute Error (MAE). The performance of the proposed method has been compared with that of other state-of-the-art methods, including the speckle reducing anisotropic diffusion, the nonlinear coherence enhance diffusion and the nonlocal means filter. The preliminary results of both simulated images and real ultrasound B-mode images indicate that the proposed method could effectively remove the speckle while preserving the edge and details.