Suppose that there are a number of alternative ways of operating a system, and a performance measure is available for comparing them. Simulation runs can be carried out to estimate this measure for different alternatives, but there are too many for all to be examined because there is a strict limit to the time available for simulations. If the accuracy with which the performance measure can be estimated increases with run length, then a balance has to be made between making few runs where the performance measure is accurately estimated and making a large number of runs but with the performance measure poorly estimated. We analyse how the best run length can be selected to ensure that an alternative is found with a good performance measure. This problem can arise in real time decision making, and we give a real example arising in the provision of fire service emergency cover.