This paper employs the multinomial logit (MNL), the dynamic multinomial logit (DMNL) and the latent class (LCM) models to analyse travellers' mode choice behaviours with consideration of preference heterogeneities in Taiwan. Based on a large-scaled nationwide survey over two consecutive years, three models are estimated and compared. Empirical results show that the DMNL model has better goodness of fit than the MNL model, suggesting the necessity of introducing inertia and slack variables into the mode choice modeling. The LCM performs significantly better than the DMNL model in terms of likelihood ratio tests, confirming the existence of heterogeneity in mode choice behaviours. The estimation results of LCM provide useful information about the heterogeneities of travellers' demographics and mode choice preferences among different segments of travellers. Accordingly, policy makers can evaluate the effectiveness of initiatives or strategies for increasing public transport usage and/or decreasing private transport usage across the segments of travellers.