In a process for combined heat and power generation there is a need for fault detection, decision support and risk assessment to prevent operational disturbances and reduction in performance. A method to achieve decision support is to use Bayesian networks, where knowledge about the process is combined with operational experience. The network covers the convectional superheaters in the flue gas train, which is a major problem domain in biomass-fuelled boilers. The superheaters are exposed to fouling from flue gases. Fouling reduces the heat transfer and result in a decreased power plant performance. The Bayesian network is constructed to give decision support on preventive action to reduce abnormal fouling. Validation of the Bayesian network show that the prediction of hard fouling works well under uncertainty.