Users have the propensity to apply real-life gender stereotypes on virtual agents. Such stereotypes could impair users' learning in virtual environments with pedagogical agents. In particular, girls and women are negatively affected by these stereotypes in math tests. The phenomenon called Stereotype Threat refers to the fear of being negatively stereotyped in a skill area that reduces learning and performance. We wonder if virtual agents can be used to reduce Stereotype Threat effects in math. For this purpose, a virtual agent has to be perceived as a social role model. We present in this paper two pilot studies with children to explore their perception of a male agent and a female agent, both child agents, weither they can establish rapport or not. We also investigate if children apply gender stereotypes when they interact with child virtual agents, and if such agents could embody an efficient role model. Results show children found the male agent as competent as the female agent, which is usually not the case in other research involving adult participants. More importantly, the child virtual agents are perceived as social role models, opening the possibility to use them to improve children performance in STEM fields.