In this paper a novel fragile watermarking scheme is proposed to detect, localize and recover malicious modifications in relational databases. In the proposed scheme, all tuples in the database are first securely divided into groups. Then watermarks are embedded and verified group-by-group independently. By using the embedded watermark, we are able to detect and localize the modification made to the database and even we recover the true data from the database modified locations. Our experimental results show that this scheme is so qualified; i.e. distortion detection and true data recovery both are performed successfully.