The purpose of this study is to examine the accuracy of ballet movement learning effectiveness on ballet beginners who have different learning characteristics through the video evaluation based on verbal feedbacks. There are few detailed hypotheses; the first, a self-modeling group assumes superior learning accomplishments than a peer-modeling group on accuracy of movement; second, a self-modeling group assumes superior learning accomplishments than a expert-modeling group on accuracy of movement; lastly, a peer-modeling group assumes superior learning accomplishments than a expert-modeling group on accuracy of movement. This study was conducted for 48 high school students whose major are dance but not ballet. 48 participants were randomly divided into 4 groups which are a self-modeling group, peer-modeling group, expert-modeling group, and control group. This study is structured by composition of 10 individual ballet movements within one minute and 30 seconds from Enchainement. In other words, this is structured in composite movement and proven results in 2 stages; acquisition and retention. The results are as below: First, self-modeling group certainly shows exceptional learning results compare to an peer-modeling group; second, self-modeling group certainly shows exceptional learning results compare to an expert-modeling group; lastly, a expert-modeling group indicates that they show limited learning results than the peer-modeling group in acquisition stage, however, there are no statistically significant between peer-modeling and expert modeling in retention stage.