Every year, more than 50,000 fires break out in Japan. Most of deceased person by fire are elderly people over 65 year old. Thus, the fire risk detection system has been needed to protect for their lives. This paper proposes the system that evaluates a degree of fire risk by watching the usage of stoves, cooking-heaters, electric appliances or monitoring atmosphere in a living-room directly and so on. As a first step of this study, we carried out laboratory experiments to monitor indoor atmosphere such as CO, H_2, and Smell using gas sensors. After the feature data is extracted from sensor output signal by the method which we proposed, the Mahalanobis' distance is evaluated for distinguishing either safety or dangerous situation.