This study employs the maximum entropy modelling technique to investigate the geographic distribution pattern of wild sheep (Ovis Orientalis) on Tangeh Sayyad Proteced Area. A set of eight environmental predictors is employed together with presence-only records of wild sheep. Two methods has been used to improve the performance of modeling: density-based occurrence thinning and performance-based predictor selection. Using the four different thresholds (Fixed cumulative value 10, 10 Percentile training presence, Minimum training presence, Equal training sensitivity and specificity), potential distribution of species was estimated. Results were evaluated using the threshold-dependent Statistics (Sensivity, Specifity, Kappa, TSS), a binomial test, Wilcoxon signed-rank test, and Area Under Curve (AUC). Relative variable importance was assessed using Maxent’s built-in Jacknife functionality. The results showed that the distributions fitted the provided occurrence data very well (at least AUCs = 0.77 for predictors with randomly selected spots and at most AUC=0.82 for random predictors with random sampling) and threshold-dependent Statistics results showed that prediction success for wild sheep were acceptable. Slope and distance to village were found to be the most important predictors. Generally, results showed that the model performance markedly improved by appropriate predictor selection and occurrence thinning.