A new blind digital watermarking algorithm based on optical Fresnel diffraction is proposed. The watermarking image is transformed into a complex matrix after discrete Fresnel diffraction. The complex matrix is separated into real and imaginary parts, which are embedded into the host image for different positions, respectively. The blind watermark is obtained by replacing the value of the embedded pixel of the original host image with the average of adjacent pixels. Numerical calculation shows that the algorithm is robust to the attack of the pixel modifications, such as brightness, contrast and gray curve, and has the resistance to the JPEG lossy compressing, cropping and superposing noise. It's also robust to the resampling attack and can be drawed from the image by the screen copy. Because the watermark can be embedded in the image flexibly, the double keys are designed in the large degrees of freedom, and the watermark can be retrieved from the published image without the original image, the algorithm has high security and practical value.