Empathic accuracy, or the degree to which one accurately understands another person's thoughts and feelings, is widely thought to be related to better social functioning, especially for women. Although previous studies have consistently found voice-only communication enhanced empathic accuracy relative to visual-only communication, there are still contradictions whether there is a difference between voice-only communication and audio-visual communication. There are 3 different assertions responding to 3 different results. The first assertion supports the prediction that voice-only communication enhanced empathic accuracy relative to audio-visual communication. Researchers support this assertion claim that accuracy is mainly the role of specific types of clues, other clues are misleading. The second assertion supports the prediction that both voice-only communication and audio-visual communication can enhance empathic accuracy because they both cotnain verbal clues. The last assertion supports the prediction that combined audio-visual communication enhanced empathic accuracy relative to voice-only communication because of more clues available in audio-visual communication. Comparing studies with different results, we found difference in inference task. Is it possible that perceivers in the empathic accuracy paradigm might direct their attention to different kinds of clues if they were asked separately about feelings versus thoughts, and accuracy might be influenced accordingly? The present paper addresses this question based on standard video paradigm of empathic accuracy in two experiments on women participants. Experiment 1 followed standard instruction condition (infer thoughts and feelings) to explore the effect of channel conditions on empathic accuracy. Three kinds of materials (full video, audio only, and silent video) were presented to 100 participants (17 - 25 years old，M = 19.15, SD = 1.47). They were asked to infer thoughts and feelings of targets in the materials. Participants wrote down their thoughts in table and estimated 25 discrete emotion words using 7-point Likert scales (0 = not at all, 6 = a great deal). Results showed that, the analysis yielded a significant main effect of channel conditions (p < .001) both in inferring thoughts and feelings. For thoughts inference, participants in full video (p < .001) and audio only (p < .001) condition exhibited higher accuracy relative to participants in silent video condition. For feelings inference, participants in audio only condition exhibited higher accuracy relative to participants in full video (p < .01) and silent video (p < .01) condition. Experiment 2 modified instruction to infer thoughts or feelings in different blocks to explore the effect of inference task on audio-visual channel effect in empathic accuracy. Full video, audio only were presented to 73 participants (17 - 23 years old，M = 19.51, SD = 1.21). Results showed that, the analysis yielded a significant interaction between channel conditions and inference task (p < .01). Simple-effects tests revealed that participants in audio only condition exhibited higher accuracy relative to participants in full video condition only for feelings inference. There was no significant difference for thoughts inference. These results suggested that voice-only communication contributed more to empathic accuracy than visual-only communication in inferring both thoughts and feelings. Whether there is a difference between voice-only communication and audio-visual communication depend on tasks to infer. Voice-only communication enhanced empathic accuracy relative to audio-visual communication only for feelings inference.