With the increased popularity of virtual worlds, hundreds of thousands of people from different physical locations can join virtual worlds. In this computer-based simulated 3D environment, avatars can both interact with each other and the environment. This new type of world has important implications for business, education, and society at large. In order to fully use the benefits of virtual worlds, it is important to know how the residents (i.e., avatars) behave, such as how they express sentiments. This research in progress seeks to study avatar sentiments in virtual worlds to examine whether and how sentiments are conveyed by avatars. Both verbal and non-verbal cues will be utilized in the sentiment analysis. To conduct the study, an advanced data collection method is leveraged to obtain various types of avatar data from a large number of real virtual world residents in Second Life in an effective and efficient way.