The blast disease of Poaceae is caused by a large species complex, among which P. oryzae is composed of several host-specialized lineages. The Pyricularia oryzae Triticum pathotype (PoT) causes the blast disease in wheat, but is also capable of infecting other grasses, which may serve as an inoculum reservoir for epidemics in wheat. In Brazil, severe wheat blast epidemics are most common in the Cerrado region. The dominant hypothesis is that signal grass (Urochloa sp.) and other gramineous plants harbor the wheat blast pathogen, thus serving as a major reservoir of inoculum for epidemics in wheat. A two-year survey of the Pyricularia blast pathogens was conducted in both wheat and non-wheat areas as well as prior (February) and during (May) the wheat growing season in Minas Gerais. A total of 1,368 plant samples representative of 31 Poaceae species, including wheat, were collected and inspected for the presence of blast symptoms. During the isolations, 932 isolates were obtained, being one fourth obtained from gramineous plants. A subset of 572 isolates was selected for identification at the species level based on portions of the CH7-BAC9 gene sequences. Most of the isolates (n = 494) were P. oryzae, within which 68% were PoT and 32% non-PoT based on two PCR assays targeting (MoT3 and C17 PCR assays). The PoT lineage was found predominantly (97%) in wheat and rarely in the other hosts, even nearby wheat fields (2.1%), as well as at longer distances from wheat regions (0.1%). The blast pathogen population isolated from signal grass grouped in different clades from PoT, and therefore referred to Urochloa lineage (PoU). A series of cross-inoculation greenhouse experiments was conducted using wheat (cv. BRS Guamirim and BR 18-Terena) and signal grass (cv. Marandu) as host and 14 PoT and six PoU isolates as pathogen factor. In the first leaf-inoculation experiment, results showed a significant interaction between host and pathogen; PoT was strongly/weakly aggressive towards wheat/signal grass and PoU was strongly/weakly aggressive towards signal grass/wheat. In inoculated wheat heads, PoT was more aggressive (>91% infected spikelets) than PoU (52% infected spikelets). In a third experiment, four signal grass cultivars (Marandu, Basilisk, Piatã, and Xaraés) were inoculated with the same set of 20 isolates. Similarly, signal grass cultivars were generally more susceptible to PoU than PoT. Severity induced by PoU was twice (7.7% severity) as high as PoT (3.8%) and so was the number of conidia/leaf produced by PoU (47,500) and PoT (23,200). Two groups of signal grass cultivars were formed, the most susceptible composed of Marandu and Basilisk and the least susceptible composed of Piatã and Xaraés. Results of our study confirm the host-specialization and the shaping of the blast populations according to the host. We further suggest that grasses in general, especially signal grass, may not play a major role as an inoculum reservoir for PoT, as it harbors mainly the PoU population. However, due to the large extent of pasture-growing regions and cross-infection ability in wheat, signal grass may harbor amounts of PoT inoculum that are sufficient for initiating leaf and head blast epidemics in wheat blast in Minas Gerais state.