Linking host spatial distributions to environmental variables can provide key information for understanding and predicting the transmission of a parasite in space. When a large number of potential intermediate hosts co-occur within a diversity of habitats, community level modelling helps to summarize such complex data set by defining groups of species/sites, i.e assemblages. We built a predictive model for the niches of small mammal assemblages including potential Echinococcus multilocularis intermediate hosts, in two areas of western China (Sichuan) by a three step modelling procedure. First, 8 assemblages were defined using a multinomial logistic model associated with a redundancy reduction procedure. Then, niches of these assemblages were modelled w.r.t. the environmental space of each sampled area using a Multiple Adaptative Regression Spline. Elevation and ETM band 6 (land surface temperature) were one of the main factors influencing assemblage distributions in both study areas. The importances of vegetation indices (NDVI and EVI) effects, which are corelated to the amount of vegetation, differed between the two locations. Finally, the model providing the lowest predictive classification error was chosen to map assemblage occurence probabilities beyond the sampled locations. We could thus discuss the predictive error component induced by extrapolation of model predictions on non trained locations.