Instructors humor is one of the effective teaching strategies. Using humor helps instructors to create a relaxing and interesting online learning climate, keep positive instructor-student relationships, and attract students to learn. In the past few decades, a plethora of empirical research has shown that humor was related to students’ academic engagement in online environment. Notably, there were several limits. First, instructor humor is a multidimensional concept, however, little is know about how different styles of humor have effects on academic engagement systematically. In addition, existing studies primarily focused on the single perspective, instructors’ perspectives (instructor humor) or students’ perspectives (perceiving instructor humor), which ignores the transmission effect of humor and the possibility of major changes in humor at the class level. Third, although instructional humor processing theory could effectively explain how instructor humor affect learning, empirical evidence still remains scarce. To address those gaps, the present studies aimed to investigate the relationships between different styles of instructors humor on students’ online academic engagement and its underlying mechanism from the individual level and the class level. One thousand eight hundred and twenty-nine college students and forty nine instructors were recruited from 22 universities of 9 provinces in China. All participants voluntarily and anonymously completed perceiving instructors humor, instructors humor, academic emotion and academic engagement scales. Meanwhile, at class level, instructors’ age, gender, seniority and size of class were collected. Students’ demographic information including gender and age were obtained. The ICCS of academic engagement at class level was .89. In addition, the present study conducted a multilevel structural equation analysis to simultaneously estimate direct and indirect effects in the case of clustered data. The results revealed that: (1) Instructors humor at class level was unrelated to students academic engagement in online learning. (2) At individual level, perceiving humor related to course content was significantly related to academic engagement, while perceiving unrelated to course content, self-disparaging and aggressive humor were irrelevant to academic engagement. (3) Academic emotion can not mediate the relationship between different styles of instructor humor and students academic engagement. (4) Positive and negative emotion had mediating effects between perceiving related to course content, unrelated to course content, aggressive humor and academic engagement. While, perceiving self-disparaging humor could affect academic engagement through negative emotion. Compared to previous research, this study further explores how different styles of humor have an influence on academic engagement in online learning. Moreover, the current study conducts a multilevel model and analyzes effects of humor from individual level and class level, which contributes to have a comprehensive understanding of how instructors’ humor facilitates academic engagement. Additionally, the present study provides empirical evidence for instructor humor processing theory and examines the mediating role of academic emotion. In practice, instructors should generate humor related to course content from students’ perspectives, and avoid unrelated to course humor, self-disparaging and aggressive humor as possible as they could.