The current study aimed at investigating how different types of embodied learnings influence elementary school students' English as a foreign language (EFL) listening performance. Two kinds of embodied learnings: real and physical body versus the 3D avatar, were compared with non-embodied learning. 69 fifth graders from two elementary schools participated in this study, and were randomly assigned into three groups (Kinect, Second Life, and paper). They learned the identical English phrases of doing sports by involving different types of embodied learnings. During the 11-week experiment, an identical EFL performance test was administered six times: before (once), during (3 times), and after (twice) the learning activities. The results depict that students learned better by watching their own 3D avatars doing motions than by moving their own bodies to produce the motions or doing nothing. Further analysis showed that the improvements made by those in the Second Life group were greater than those made by the participants in the other two groups when the performances of students with low achievement were compared. It was also found that learning by watching one self's avatar benefits both students with high- and low-achievement in EFL performance. Some suggestions for pedagogical applications and future research are also provided.