In this paper we present a new class of collision attacks that are based on inducing faults into the encryption process. We combine the classical fault attack of Biham and Shamir with the concept of collision attacks of Schramm et al. Unlike previous fault attacks by Blomer and Seifert our new attacks only need bit flips not bit resets. Furthermore, the new attacks do not need the faulty ciphertext to derive the secret key. We only need the weaker information whether a collision has occurred or not. This is an improvement over previous attacks presented for example by Dusart, Letourneux and Vivolo, Giraud, Chen and Yen or Piret and Quisquater. As it turns out the new attacks are very powerful even against sophisticated countermeasures like error detection and memory encryption.