Flood is one of the worst natural disasters in New South Wales (NSW) Australia. To reduce damage from flood events, precise estimates of floods are needed to design flood-safe infrastructure. In this regard, a risk-based approach is used where design flood or flood quantile is adopted such as 100-year flood. To estimate flood quantiles, flood frequency analysis is generally adopted, which however can only be applied when a long period of flood data is available at the site of interest. However, at many locations, long or quality flood data is unavailable. In this case, a regional flood frequency analysis (RFFA) is adopted. In RFFA, identification of homogeneous regions is a primary step. Multivariate statistical techniques such as cluster analysis can be used to find homogeneous regions. In this study, we use data from 176 NSW catchments to identify homogeneous regions for RFFA by cluster analysis. The median relative error values from the two groups, formed by cluster analysis, are smaller than that of the combined dataset. It has been found that cluster analysis can reduce the model error in RFFA.