In this paper, a robust watermarking scheme operating in curvelet domain is proposed. Curvelet transform directly take the edge as the basic representation element; it provides optimally sparse representations of objects along a general curve with bounded curvature. The image is partitioned into small blocks, Unequally Spaced Fast Fourier Transforms (USFFT) is employed to decompose each block into curvelet domain. We embedded the watermark into the selected blocks, scale and curvelet coefficients based on the edge map of cover image. As usual, the watermarks are blindly detected using a correlation detector. Experimental results demonstrate that the embedded watermark survives severe image attacks and shows advantages over watermark algorithm in the wavelet domain. Both the robustness and the transparency can be well proved by the proposed watermarking algorithm.