The central research question of this empirical study was: How do student demographics, math self-efficacy, and math anxiety relate to and predict their choice of pedagogical agents serving as virtual math tutors? A total of 152 middle school students and 135 college students were surveyed on their perceived math self-efficacy, math anxiety, and their choice of virtual math tutors. The collected survey data were analyzed using chi-square tests and logistic regressions to answer the research question. Results suggest that (1) student gender predicted choice of agent gender for both groups of samples; (2) student ethnicity predicted choice of agent ethnicity for the middle school students but not for the college students; and (3) student math self-efficacy predicted choice of agent gender for the middle school students and choice of agent age for the college students. Overall, the results supported the similarity-attraction theory in terms of gender with both groups of samples in their agent choice. However, similarity in ethnicity seemed a more important factor for the middle school participants than for the college participants in their agent choice. The results also indicate that students' agent choice might be influenced by their perceptions of their own self-efficacy level and the perceived agents' self-efficacy level. Future research should be conducted to verify the results.