The problems of harbour security and safety at sea have gained attention in recent years due to an increase in the threats posed not only by military actions but also terrorism and criminal activities. For harbours and harbour approaches, there is a need for underwater surveillance systems. We have studied techniques for how underwater surveillance could be optimized for best performance at a given cost. We report on the general principles of such optimizations. Applying these to the particular problem of finding an optimal sensor configuration, we find it very difficult to set up an optimization scheme that would give the solution in one go. Instead the problem should be broken down into subproblems, which can be solved. By means of comparing different sensor configurations with efficient visualisation tools, a solution can then be found manually. An important subproblem is harbour surveillance with active sonar, where the performance is determined by the underwater environment. In a harbour there are many false echoes (reverberation), which are difficult to model using common 2-D models. We find that better results are obtained with a 3-D model, which has been used for the simulations reported here. Our conclusions are that existing sensor support systems need to be extended with 3-D models and good visualisation tools.