The occurrence probabilities of heavy rainfalls brought about flood events have an essential role in designing water-related structures and water resource management. In many cases, data for analysis are either not available or are insufficient for reliable design of water-related structures. Regional frequency analysis is frequently preferred to provide design information at sites with especially inadequate data available. Our study implemented L-moment procedures to annual maximum rainfall series from 70 gauging stations in the Middle Black Sea Region (MBSR) in Turkey to estimate regional rainfall quantiles. The first attempt for regionalization was to evaluate the entire area as an only homogeneous region. The sub-regions were initially defined with the ward’s clustering algorithm due to the presence of discordant sites under a presumption of a single homogeneous region. In compliance with the results of the discordancy and heterogeneity measures, the most promising classification was achieved with 6 clusters (sub-regions). The sub-regions have satisfied the homogeneity condition as “acceptably homogeneous.” The GEV and GLO distributions in five sub-regions, GNO distribution in four sub-regions, and PE3 distribution in three sub-regions were decided to be acceptable as a regional frequency distribution. In comparison, GPA is a candidate distribution in none of six sub-regions.