A multiresolution edge detection algorithm for speckle images is proposed. Due to the signal dependence of speckle noise, the vanance of a speckle image depends on the local average intensity; thus an edge detection method independent of the local average intensity is desirable for correct extraction of real, significant changes in an original signal. In the proposed method, each area having different resolution is first classified according to the statistical properties of a speckle image, namely, a discontinuity measure such as the ratio of variance to mean square or the maximum difference between the real and theoretical cumulative density functions. Then the real edges are extracted in a multiresolution environment. Computer simulation with several test images shows that the proposed method significantly reduces false edges in relatively homogeneous areas while detecting fine details properly. Also, simulation results from the conventional edge detection methods for speckle images are compared with those of the proposed method.