The present paper describes the development of neural network based controllers (NNBC) to control the outlet hot water temperature from a chain grate, stoker fired boiler while maintaining the oxygen level in the exhaust within target values. The controller initially matches the fuel input to the boiler to the error between the measured and set point water temperatures. In one version of the system a neural network then uses the measured current oxygen and carbon monoxide levels in the boiler exhaust to adjust the combustion airflow. In the second version two artificial neural networks were used to estimate the concentrations of the exhaust oxygen and carbon monoxide 3 min into the future. This further network was able to characterise the dynamics of the combustion process and the 'future' predictions were in good agreement with the plant data collected on the industrial chain grate stoker. Consequently the predicted values were then used by the first network to control the combustion airflow and hence maintain the oxygen level within specified limits. Test results showed that both versions of the novel NNBC were able to control the hot water outlet temperature while keeping the excess air level within the desired levels, and in so doing, the NNBC also managed to maintain low pollutant emissions.