A transition in robotics from tools to teammates is underway, but, because it is in an early state, experience with intelligent robots and agents is limited. As such, human mental models of intelligent robots are primitive, easily influenced by superficial characteristics, and often incomplete or inaccurate. This paper investigates the factors that influence mental models of robots, and explores solutions for the formation of accurate and useful mental models with a specific focus on military applications. Humans must possess a clear and accurate understanding of how robots communicate and operate, particularly in military settings where intelligent, autonomous robotic agents are desired. Complete and accurate mental models in these hazardous and critical applications will reduce the inherent danger of automation disuse or misuse. Implications for training and developing appropriate trust are also discussed.