This paper presents two attacks on Teng et al.'s fragile watermarking algorithm. Both attacks allow the attacker to apply valid watermarks on tampered images, therefore rendering the watermarking scheme useless. The first attack uses the watermarked version of two chosen images, and the second attack, a generalization of the first, uses a number of arbitrary watermarked images. The paper also models the cryptanalysis process for the second attack using Markov chains in order to demonstrate that the necessary number of images is relatively small for a high probability of successful attack. All the results that are presented in this paper have been confirmed by a practical implementation.