Abstract: Aim To study the decision rules in the structure of multiple level decision fusion and examine the feasibility of these decision rules. Methods The decision rules for fusion units in every level were provided by use of the minim- um Bayesian risk criterion. Different loss functions were used to discuss their effect on the system performance. Results Different ratios of the loss of false alarm to the loss of non- detection resulting in different system performance. Compared to N- P criterion, Bayesian method can make use of the prior knowledge and the sample knowledge. Conclusion The system performance become more sound and it is easier to calculate the thresholds of the decison rules. Bayesian method has great flexibility in dealing with the practical problems about decision fusion.