Wheat blast, caused by Pyricularia oryzae Triticum (PoT), is an emerging threat to global wheat production. The current understanding of the population biology of the pathogen and epidemiology of the disease has been based on phylogenomic studies that compared the wheat blast pathogen with isolates collected from grasses that were invasive to Brazilian wheat fields. In this study, we performed a comprehensive sampling of blast lesions in wheat crops and endemic grasses found in and away from wheat fields in Minas Gerais. A total of 1,368 diseased samples were collected (976 leaves of wheat and grasses and 392 wheat heads), which yielded a working collection of 564 Pyricularia isolates. We show that, contrary to earlier implications, PoT was rarely found on endemic grasses, and, conversely, members of grass-adapted lineages were rarely found on wheat. Instead, most lineages were host-specialized, with constituent isolates usually grouping according to their host of origin. With regard to the dominant role proposed for signalgrass in wheat blast epidemiology, we found only one PoT member in 67 isolates collected from signalgrass grown away from wheat fields and only three members of Urochloa-adapted lineages among hundreds of isolates from wheat. Cross-inoculation assays on wheat and a signalgrass used in pastures (U. brizantha) suggested that the limited cross-infection observed in the field may be due to innate compatibility differences. Whether or not the observed level of cross-infection would be sufficient to provide an inoculum reservoir, or serve as a bridge between wheat growing regions, is questionable and, therefore, deserves further investigation.