Those existing robust watermarking algorithm in singular value decomposition domain always inserts an watermark into singular values from singular value matrix.A robust zero-watermarking algorithm based on variance was proposed, using two orthogonal matrices deriving from singular value decomposition. The image was split into non-overlapping blocks, and each block was transformed with singular value decomposition. The zero-watermark sequence was produced by judging the numerical relationship between variance of each element′s square from the first column of U matrix and V matrix. It has perfect invisibility, since no modification has been made to the original image. Its robustness towards resampling, smoothing, noise adding and JPEG compression is tested. The experimental results show that it has strong robustness towards those above attacks.