In this paper, we consider the failure propagation in a power system which is coupled with a cyber network. We identify the ratio of the infection rate of malware to the tripping rate of elements in the power network, defined as cyber-physical propagation ratio (CPPR), as a crucial parameter, and show that CPPR affects the propagation pattern. When CPPR is very small, the overloading effect in the power system dominates, and the failure propagation profile shows a typical 'jump' or step pattern. Moreover, as CPPR increases, the step magnitude rapidly reduces and becomes less noticeable. Furthermore, we develop a simple approach to distinguish various propagation patterns for different values of CPPR. Results show that CPPR effectively characterizes the significance of cyber coupling and hence the relative impact of malicious attack from the cyber network, and its value determines how cascading failure propagates in a cyber-coupled power network.