As a result of dazzling advancement of web technology, e-learning system has emerged as a very effective venue with which learners can interact with online contents to get a full understanding on target subjects. Despite such success, empirical studies on e-learning performance lack rigorous theory with which learners’ e-learning participation intention is analyzed. In this sense, we propose an extended transactional distance theory (ETDT). Transactional distance (TD) is the physical separation that leads to a psychological and communications gap, a space of potential misunderstanding between the inputs of the instructor and those of the learner. This study focuses on e-learning environment where avatar instructor is used. To prove the validity of the ETDT, we assumed that dialogue is represented by avatar types (attractive vs. expert) and structure by contents difficulty (easy vs. difficult), and that TD_PI is represented in a reverse proportion to path coefficient from either avatar trust (AT) or contents trust (CT) and learners’ PI (Participation Intention). Results from experiment with valid 205 questionnaire revealed that (1) both TDAT_PI and TDCT_PI decrease when dialogue is administered by expert avatar and structure is composed of easy contents, and that (2) TDCT_PI shows no difference between easy contents and difficult contents when dialogue is administered by attractive avatar.