Accurate estimation of grazing capacity is critical for sustainable rangeland management, yet remains challenging in semi-arid systems due to spatial heterogeneity and data uncertainty. This study applied an Adaptive Neuro-Fuzzy Inference System (ANFIS) to predict grazing capacity in semi-steppe rangelands of northwest Iran and eastern Turkey. Four ecologically relevant inputs-slope, forage production, water supply distance, and soil resistance to erosion-were used as predictors. Forage production was derived from NDVI-biomass calibration (R2 = 0.72, RMSE = 58.3 kg/ha), and unsuitable areas (slope > 60 %, biomass < 50 kg/ha) were excluded. The ANFIS model was implemented in MATLAB using Gaussian membership functions (three per input), a cluster radius of 0.35, and 16 fuzzy rules. Model evaluation showed strong performance on training data (NRMSE = 4.7 %) but a substantially higher error on testing data (NRMSE =19.2 %), indicating potential overfitting and spatial heterogeneity effects. Spatial outputs classified the study area into five grazing capacity categories, with higher capacities in western and southern zones and lower capacities in central regions. Comparison with vegetationtype classifications highlighted differences arising from ANFIS's integration of multiple drivers beyond forage biomass. While results demonstrate the promise of neuro-fuzzy approaches for handling uncertain datasets and capturing spatial variability, we emphasize that outputs should be interpreted as indicative patterns rather than prescriptive management recommendations. Future work should integrate field validation, benchmark against simpler models, and incorporate dynamic factors such as drought and livestock species differences to enhance ecological realism.