The invention relates to a fire detection method using flame color templates. Adopting a colorful CCD photography system, based on the regularity of distribution of the flame three primary colors in the color space, the invention collects a plurality of flame samples beforehand, makes analytical statistics on the flame color values to obtain sample average and variance, and inputs statistical results into computers as parameters. When the detector is working, the colorful CCD converts on-site images into digital signals, and sends to the computer internal memory through an image card; after obtaining the image data, the processor compares the data with the predetermined parameters, to judge whether the data conforms to the Gaussian distribution law; regions which meet the conditions are taken as questionable regions, and then the questionable regions are under differential processing to further analyze whether changes of the objects over time meet the characteristics of fires such as jump, flicker and instability. The invention increases reliability of fire detection, and can meet different requirements of fire detection, linked alarming, image monitoring, as well as fire and theft prevention.