When selecting method of flame′s colors and spatial-temporal features in general illumination in order to find the best feature adapting to the detection of the flame area, PCA feature selection method is used to reduce characteristics of each color channel in different color characteristics. The Relief feature selection method is used to determine the optimal classification for these candidate color and moving subsets. Moreover, contribution ratio of discrimination of all selected colors and moving features is validated by covariance matrix. In the course of the experiment, three sets of fusion features are used for analysis. Experimental results show that the flame detection system based on selected color and moving feature has higher accuracy and recognition efficacy.