As emotion provides an important clue for communication, the computer needs to be more sympathetic to users’ commands in the context of their emotion. A challenging attempt has been made to develop an emotional computer, which reads such physiological signals as photoplethysmogram, electrodermal activity and skin temperature and analyzes them online with a rule base into human emotion. We adopted a two-dimensional emotion model and a number of empirical studies have been conducted to find out some valid physiological parameters and to map them with nine categories of human emotions. Some research challenges were found that need to be addressed until the emotional computer comes to the market.