The implementation of travel time reliability (TTR) in route choice behaviour is still not very common in transport models, especially not in a public transport context. The reasons probably are that it is difficult to measure and that there is no agreement how it best can be represented in utility functions. Typically, it is represented by a standard deviation, however, particularly in public transport choices it is more likely that travellers think about the consequences of unreliability in travel times in terms of buffer times. This paper contributes to the literature by comparing five different model specifications of TTR in public transport route choices that are either based on standard deviations or on buffer time indicators. The models are estimated from choices observed in a stated choice experiment. To address heterogeneity, a latent class model is estimated. The results suggest that the reliability buffer time indicator outperforms the standard deviation indicator. Furthermore, the reliability buffer time parameter is only statistically significant in two of the four classes. The other two classes are particularly sensitive to making transfers and to low frequencies of public transport services, suggesting different strategies to deal with TTR.