Watermark robustness to various kinds of attacks is still an problem. Usually, single watermark cannot guarantee enough robustness when several attacks are encountered at the same time. In order to solve the problem, a new multi-image watermarking algorithm is proposed in Discrete Cosine Transform (DCT) domain. The original image is first sub-sampled, and the sub-images are separated into two groups. Then binary watermark images are embedded by utilizing two kinds of modulations, i.e. dither modulation and DCT coefficient modulation based on selected blocks. Compared with existing methods, the proposed algorithm can obtain better image quality. As the watermark images are embedded at different frequency components, at least one watermark can survive common attacks, such as JPEG compression, wiener filtering, brightness and contrast enhancement. Experimental results show that the proposed scheme is efficient to resist traditional signal processing attacks. At least one watermark can survive multi-attacks, which provide proof for copyright protection.