Linear programming methods and non-linear, evolutionary algorithm-based optimization techniques have been shown to be effective in managing large-scale air traffic flow problems. However, many of these algorithms assume perfect knowledge therefore the robustness of these algorithms in the presence of uncertainties is questionable. Since real-world application of these methods require them to be effective under uncertainty (i.e. produce few unexpected capacity violations), it is critical that they are tested in such conditions. In this paper we test the effectiveness in the presence of uncertainly of a binary programming approach and a novel, fast-learning evolutionary algorithm. Specifically we change the assumed takeoff times on which these algorithms are trained, and test the resulting solutions when takeoff delays that are consistent with historical data are incorporated. Experimental results show that without uncertainly, both sets of algorithms are able to quickly produce solutions with few to no violations. In the presence of uncertainty, the performance of the algorithms degrade with respect to the amount of delay added, but are still very good. Even when uncertainty is extremely high, the expected delay is never increased more than 30%.