Several models of route choice in frequency-based assignment are compared for underlying assumptions on service regularity, passenger information, and choice set structure. Numerical results for some simple examples show that route splits differ significantly under different assumptions, so for practical applications the selection of the most suitable choice model is important and none of the models can be regarded as a good approximation for all possible assumptions. Sensitivity of route choice against perturbations of running times or service frequencies is another consideration, because a continuous response improves convergence in demand models with feedback. Finally, it is demonstrated that in terms of expected travel time, the decision about when to alight (and where to continue the journey) is just as important as the decision of which line to board.