To meet the watermarking requirement in encrypted domain, a novel scheme for robust and separable watermarking in encrypted image is proposed based on Compressive Sensing (CS). Firstly, the content owner divides the original image into non-overlapping blocks, and then the edge-detection method is utilized to classify all blocks into significant or insignificant blocks. For the former, traditional method is used for encryption; and for the latter, CS is used for encryption, which leaves some space for embedding data. Then, the binary watermark is permutated with the data hiding key, and embedded into the encrypted image. The way to obtain the image content and watermark is separable, and the attributes of the block can be regained according to pixel distribution of the watermarked image, which avoids transmitting the attribute information. Furthermore, the watermark is embedded four times in the encrypted image, which guarantees its robustness. The experimental results show that the proposed scheme is robust and secure against moderate attacks.