The CIMMYT germplasm collection contains 22,000 samples of maize. Accessing this diversity for practical plant improvement has been hampered by sheer numbers, lack of information on breeding value and traits contained in each accession, and the heterogeneous nature of many of the accessions. We have developed a rapid SSR assay for inbred lines and another for populations, which are screened as bulks to improve efficiency. Maize landraces contain a high number of alleles and a very high gene differentiation (Gst), indicating a highly structured pattern of diversity, but ~ 90% of the diversity is still found within, and not between landraces. CIMMYT OPVs suffered a tremendous decline in Gst, indicating a high degree of mixing and few generations to differentiate populations; however, total allelic diversity is actually higher than in the landraces. Inbred lines from CIMMYT contain yet more diversity than temperate inbreds, CIMMYT OPVs, and Mexican landraces, because they were selected from breeding material containing a highly diverse background, but they completely lack clear groupings. When 1,518 landraces, OPVs, and inbred lines from CIMMYT, Africa, and Asia were characterized in the Generation Challenge Program, clear structure was observed based on geographic region, altitude (in the case of landraces), breeding program and pedigree (in the case of improved materials). A core of 234 inbreds and 150 individuals from 30 populations was chosen based on the markers, in order to maximize the genetic distances within the core. Measures of genetic diversity and distance are as high in the core as they are in the entire set. This core will be used for association mapping studies, starting with a pilot study searching for association between vitamin A content and genes from the carotenoid pathway, and with luck ending with the determination of several genes accounting for increased drought tolerance in maize germplasm.