In this paper, flame image processing technique is used to estimate NOx emissive concentration of the pulverized coal boiler. According to the brightness value of flame image pixels, the temperature distribution of the flame is calculated. The combustion characteristic parameters are picked up from the flame image. Based on the back propagation neural network, the model to estimate the NO, concentration in the gas released from the power plant boiler has been developed. The experiments on the furnace flame monitor system of a power plant have testified that this estimation model can get the satisfying results. (C) 2002 Elsevier Science Ltd All rights reserved.