The assessment of flame combustion stability play a vital role in the safe, economic, environment protection operation of the large capacity pulverized coal boilers. In this paper, the flame radiation intensity signal is acquired by a low-cost flame detection device. Fast Fourier transform is used to exact power spectrum of frequency domain and phase domain from time domain signals based upon wavelet de-noising methodology. The characteristic parameters??? the variance of the intensity value, the flicker frequency and the average phase, which reflect time domain, frequency domain and phase domain respectively, are extracted. A compositive assessment model for the stability of flame combustion using the grey theory is proposed which is based on the extraction of flame radiation characteristic parameters from measured data, and a stability index system for the assessment of flame combustion established. The assessment results demonstrate this method is feasible and efficient. As a result, a promising approach is introduced for the diagnose of the flame stability.