In 2010, Zhao et al. proposed a fault-tolerant group key agreement protocol based on RSA cryptosystem. In their protocol, malicious participants can be detected in the fault detection phase. Thus, they do not disrupt the key agreement process. However, there exists a drawback in the protocol of Zhao et al. After two rounds of trying, an adversary can impersonate any other group member in the third round. We propose an improved protocol to overcome this drawback. The efficiency and security analysis of our improved protocol are also provided in detail. Copyright © 2012 John Wiley & Sons, Ltd.