Multimedia pedagogical agents are on-screen characters that allow users to navigate or learn in multimedia environments. Several agents' characteristics may moderate their instructional effectiveness, including appearance, gender, nonverbal communication, motion, and voice. Here, we conducted a meta-analysis to test hypotheses from diverse theories predicting the effects of these agents' characteristics. We tested predictions of cognitive load theory, cognitive theory of multimedia learning, computers are social actors, social agency theory, uncanny valley, and the action observation network. Our meta-analysis of 32 effect sizes (N = 2104) revealed a small overall effect (g+ = 0.20), showing that learning with multimedia pedagogical agents was more effective than learning without these agents. As predicted by the redundancy effect of cognitive load theory and the coherence principle of cognitive theory of multimedia learning, 2D agents (g+ = 0.38) tended to be more effective than 3D agents (g+ = 0.11). As predicted by the computers are social actors hypothesis, most of the agents' characteristics, including nonverbal communication, motion, and voice, appeared not to moderate their effectiveness. We conclude that multimedia pedagogical agents help learning through multimedia, and that students may be able to learn similarly from different types of agents.