A novel two-phase robust content-based image authentication scheme is introduced. The proposed scheme is constructed based on a combination of hard and soft authentication using two existing generic approximate message authentication codes (AMACs). The AMACs combine error- correcting codes with cryptographic primitives such as message authentication codes and symmetric encryption algorithms. The message authentication codes are used for hard authentication, whereas the error-correcting codes introduce a certain degree of robustness in authentication. This is achieved by correcting minor unintentional modifications as a result of common image processing operations such as quantization , compression , and noise addition. The two-phase image authentication scheme verifies the authenticity of an image in two phases. The low frequency elements of the image in a transform domain are subjected to the first phase while some higher frequency elements are left to the second phase if the first phase succeeds. The proposed scheme tolerates common content- preserving modifications in an image but can discriminate intentional modifications affecting the image content. Mathematical bounds for the accuracy and the security level of the proposed approach are estimated and the performance is compared with some other well-known schemes in the literature. The results demonstrate that the proposed scheme shows high discriminating capability and can detect different types of meaningful forgery attacks on images while preserving the robustness. It also outperforms the benchmark image authentication schemes in terms of tradeoff between robustness and fragility.