Abstract: Bayes' Theory of conditional probability and efficient algorithms for probability computation together allow a probabilistic approach to expert systems known as Bayesian Belief Networks (BBNs). BBNs facilitate the graphical representation of complex problems and allow users to make expert predictions on the likelihood of a hypothesis in the absence of complete information. As such they seem applicable to the problem solving in the face of uncertainty that characterises enemy course of action (COA) assessment at the tactical level of war. In this paper, BBNs were constructed to reflect two distinct tactical intelligence problems, one based on conventional operations, and one based on peace support operations (PSO). Difficulties were encountered in quantifying the PSO BBN because the intelligence collection plan reflected a requirement to collect information on enemy capabilities as well as intent. Consequently, the BBN had to be modified. Overall, however, it was found that BBNs could be constructed to reflect the tactical enemy COA assessment problem. Nevertheless, it was found that the utility of such BBNs was limited, especially in the conventional environment, because of the likely requirement to modify quantification to reflect actual battlefield factors such as weather and terrain, even for the same set of COA. It was considered that the development of a library of BBN fragments prior to deployment could go some way to alleviate the problem, although mainly in the PSO environment with a slower operational tempo. On the other hand, the modification problem could be solved by making the BBN more general, allowing it to be used as a tactical indicator and warning tool, at least in the PSO environment.