Affective communication, communicating with emotion, during face-to-face communication is critical for social interaction. Advances in artificial intelligence have made it essential to develop affective human-virtual agent communication. A person's belief during human-virtual agent interaction that the agent is a computer program affects social-cognitive processes. Whether this belief interferes with affective communication is an open question. We hypothesized that the imitation of a positive emotional expression by a virtual agent induces a positive emotion, regardless of the belief. To test this hypothesis, we conducted an fMRI study with 39 healthy volunteers, who were made to believe that a virtual agent was either a person or a computer. They were instructed to smile, and immediately afterwards, the virtual agent displayed a positive, negative, or neutral expression. The participants reported a positive emotion only when their smile was imitated by the agent's positive expression regardless of their belief. This imitation activated the participants' medial prefrontal cortex and precuneus, which are involved in anthropomorphism and contingency, respectively. These results suggest that a positive congruent response by a virtual agent can overcome the effect of believing that the agent is a computer program and thus contribute to achieving affective human-virtual agent communication.