AbstractResource selection functions (RSFs) are widely used approach of mathematical modelling for the analysis of presence-absence data to deduce wildlife-habitat relationships. However, the generation of RSFs may be hampered by a wild animal’s population size and density. Consequently, the prediction accuracy of RSFs for low density species is disputed. We examined the validity of RSFs in a study of the Tibetan fox (Vulpes fer-rilata), a low density species, in Shiqu County, Western Sichuan Province, China. Our RSFs model was con-structed based on V. ferrilata habitat data collected from 2001 to 2003. Six environmental variables were considered with reference to V. ferrilata den habitat: water distance, vegetation type, pika (Ochotona sp.) den quantity, den location on slope, gradient, and aspect. In order to examine the validity of our RSFs model, we used a principal components analysis (PCA) to re-analyze our data. The total accuracy of our RSFs was determined to be 75.2%. The AIC value was 309.172. A ROC curve revealed, when the sensitivity was 0.857, the specificity of our RSF model was 0.353. The Nagelkerke R² was 0.485. Only three variables were judged as important by our RSFs: aspect, position on the slope and vegetation type, of which aspect was the most informative variable of the three. However, partial correlation coefficients of the three variables were very low. These results revealed that the RSFs model to predict den habitat for V. ferrilata did not fit well. Considering the result of PCA, the importance ranks were (in order of decreasing importance): pika den quantity, water distance, gradient, position on the slope, vegetation type, and aspect. We offer ex-planations as to why, for this species, a RSFs model may not be generated accurately and why the results of the two analyses were so different. We suggest other analytic methods that may be combined to yield a more comprehensive result for wild animal species with low population sizes and densities.