Zoonoses are an important class of infectious diseases. An important element determining the impact of a zoonosis on domestic animal and human health is host range. Although for particular zoonoses some host species have been identified, until recently there have been no methods to predict those species most likely to be hosts or their relative importance. Complex inference networks infer potential biotic interactions between species using their degree of geographic co-occurrence, and have been posited as a potential tool for predicting disease hosts. Here we present the results of an interdisciplinary, empirical study to validate a model based on such networks for predicting hosts of Leishmania (L.) mexicana in Mexico. Using systematic sampling to validate the model predictions we identified 22 new species of host (34% of all species collected) with the probability to be a host strongly dependent on the probability of co-occurrence of vector and host. The results confirm that Leishmania (L.) mexicana is a generalist parasite but with a much wider host range than was previously thought. These results substantially change the geographic risk profile for Leishmaniasis and provide insights for the design of more efficient surveillance measures and a better understanding of potential dispersal scenarios.