The training data yielded models that predicted the distribution of test points significantly better than random for bothspecies. For Heteromys australis, the preliminary modelpredicted potential presence for the species in 20.4% of themap pixels of land area. Of the 14 test points, 12 fell in pixelsof predicted presence, and two lay in pixels of predictedabsence (binomial probability, P < 0.0001). The two that fellin predicted absence were within 3 km of areas of predictedpresence. Similarly, the preliminary model for H. anomaluspredicted presence in 22.0% of the map pixels. Five of the10 test points were located in pixels of predicted presence(binomial probability, P = 0.0482), and the other five pointslay within 10 km of areas of predicted presence