Combining morphological and molecular data to identify genetic variation and marker–trait association is one of the most important prerequisites for genomics-assisted selection in crop improvement. To this end, a total of 138 finger millet (Eleusine coracana subsp. coracana) accessions including five improved varieties were evaluated to assess the genetic variation and population structure and undertake association mapping. These accessions were basically collected from Ethiopia (96), Eritrea (8), Kenya (7), Zambia (9) and Zimbabwe (13). Finger millet accessions were evaluated in the field for 10 important agronomic traits and also characterised using a set of 20 microsatellite markers. Mean polymorphism information content of 0.61 was observed from a total of 222 alleles with an average of 11.1 alleles per microsatellite locus. About 61% of alleles detected were rare (<5%) and specific allele amplification was observed in 34 accessions. Both weighted neighbour-joining based clustering using molecular data and hierarchical clustering using phenotypic trait data grouped the 138 accessions into four major clusters that were not entirely based on their geographical origins. Genome-wide association studies depicted 16 significant (P < 0.01) associations between 13 microsatellite markers and six agronomic traits. Our results reveal a unique abundance of rare alleles in finger millet and highlight the need for more careful selection of genome-wide association studies in the future in order to capture the contribution of rare alleles to important agronomic traits.