Regional frequency analysis (RFA) aims to estimate quantiles of extreme hydrological variables (e.g. floods or low-flows) at sites where little or no hydrological data is available. This information is of interest for the optimal planning and management of water resources. A number of regional estimation models are evaluated and compared in this study and then used for regional estimation of flood quantiles at ungauged catchments located in the Montreal region in southern Quebec, Canada. In this study, two neighborhood approaches using canonical correlation analysis (CCA) and the region of influence (ROI) method are applied to delineate homogenous regions. Three regression methods namely log-linear regression model (LLRM), generalized additive models (GAM), and multivariate adaptive regression splines (MARS), recently introduced in the RFA context, are considered for regional estimation. These models are also applied considering all stations (ALL). The considered models, especially MARS, have never been used previously in a concrete application. Results indicate that MARS and GAM have comparable predictive performances, especially when applied with the whole dataset. Results also show that MARS used in combination with the CCA approach provide improved performances compared to all considered regional approaches. This may reflect the flexibility of the combination of these two approaches, their robustness, and their ability to better reproduce the hydrological phenomena, especially in real-world conditions when limited data are available.