Habitat fragmentation is a hot issue of global biodiversity conservation, and corridor construction is one of the important strategies to reduce the negative impact of habitat fragmentation. To date, many models have been used for the prediction location of animal corridors, one key concern from conservation biologist is whether these models can accurately predict actual dispersal paths of target species. In this work, the white-headed langur (Trachypithecus leucocephalus), an unique endangered primate to Guangxi, China, was chosen as target species, it inhabits karstic rocky mountain areas, where is facing severe fragmentation problems due to increased human disturbance. During the last century, this species experienced a severe population declines, with an estimated 80% contraction in its distribution and a 60% reduction in its total population. They are now only distributed in small isolated habitat patches with limited opportunities to exchange gene among populations. The corridors can help this species disperse from one patch to another, and the connectivity of isolated patches is enhanced consequently. Least Cost Path (LCP) model and Conditional Minimum Transit Cost (CMTC) model are two commonly used corridor site selection models, whereas their outputs have never been validated using the actual observations data. For this purpose, LCP model and CMTC model were used to simulate the dispersal corridors of T. leucocephalus, respectively, and the simulation results from two models were further compared with field observated data for the accuracy test of these two models. The results revealed that:The fully accuracy rate for predictions of LCP model was 46.7%, partial accuracy was 20%, the complete inaccuracy rate was 33.3%. The full accuracy rate for predictions of CMTC model was 26.7%, the remnant 73.3% were all partially accurate and there was no completely inaccurate outputs from CMTC model. In general, the accuracy rate for predictions of CMTC model was higher than the outcome of LCP model, and thus the CMTC model had better performance on prediction for corridor location of T. leucocephalus than LCP model. The type of input source feature, grid size configuration of cost surface, changes in habitat land use types and complexity of animal movement behavior were four main factors affecting the accuracy of model prediction results.