Whereas fisheries acoustics data processing mainly focused on the detection, characterization and recognition of individual fish schools, we here addressed the characterization and discrimination of fish school clusters. The proposed scheme relied on the application of the Bags-of-Features (BoF) approach to acoustic echograms. This approach is widely exploited for pattern recognition issues and naturally applies here considering fish schools as the relevant elementary objects. It relies on the extraction and categorization of fish schools in fisheries acoustic data. Echogram descriptors were computed per unit echogram length as the numbers of schools, in different school categories. We applied this approach to the discrimination of juvenile and adult anchovy (Engraulis ringens) off Peru. Whereas the discrimination of individual schools is low (below 70%), the proposed BoF scheme achieved between 89% and 92% of correct classification of juvenile and adult echograms for different survey datasets, and significantly outperformed classical school-based echogram characteristics (about 10% of improvement of the correct classification rate). We further illustrate the potential of the proposed scheme for the estimation of the spatial distribution of juvenile and adult anchovy populations.