Estimating the minimum streamflows in rivers is essential to solving problems related to water resources. In gauged watersheds, this task is relatively easy. However, the spatial and temporal insufficiency of gauged watercourses in Brazil makes researchers rely on the hydrological regionalization technique. This study's objective was to compare different hierarchical and non-hierarchical clustering approaches for the delimitation of hydrologically homogeneous regions in the state of Rio Grande do Sul, Brazil, aiming to regionalize the minimum streamflow that is equaled or exceeded in 90% of the time (Q90). The methodological development for the regionalization of Q90 consisted of using regression analysis supported by multivariate statistics. With respect to independent variables for regionalization, this study considered the morphoclimatic attributes of 100 watersheds located in southern Brazil. The results of this study highlighted that: (i) the clustering techniques had the potential to define hydrologically homogeneous regions, in the context of Q90 in the Rio Grande do Sul State, mostly the Ward algorithm associated with the Manhattan distance; (ii) drainage area, perimeter, centroids X and Y, and mean annual total rainfall aggregated important information that increased the accuracy of the cluster; and (iii) the refined mathematical models provided excellent performance and can be used to estimate Q90 in ungauged rivers.