To reduce flood risk, it is crucial to accurately estimate design floods, which is a flood discharge associated with a specific annual exceedance probability. The main aim of this study is to devise an approach to carry out a Regional Flood Frequency Analysis within hydrologically similar sub-regions to estimate design floods for gauged and ungauged catchments with diverse climate and catchment characteristics. In such catchments, traditional regionalisation approaches usually fail due to a high level of inhomogeneity. The developed RFFA approach was evaluated using 363 catchments in New Zealand, a country with hydrologically diverse regions. It was found that using climate zones and catchment characteristics led to a higher degree of homogeneity than traditional sub-divisions. Cluster analysis based on catchment attributes was applied to further delineate homogenous regions, which resulted in 21 sub-regions. The two-parameter Log-Normal and Pearson 3 distributions were identified as the dominant regional probability distributions for these sub-regions. Next, the Generalised Additive Model coupled with the Index Flood L-moment approach was employed to estimate regionalised design floods of various return periods. Model performance was assessed using a Jackknife Resampling procedure. The results indicated significantly smaller error estimates for all estimated design floods than prior RFFA studies in New Zealand. The approach provides region-specific design values to inform flood risk management. It allows for robust design flood estimation in both gauged and ungauged catchments and can be implemented in other regions.