Extended Abstract Background: Soil organic carbon is one of the important parameters to determine soil fertility, and production ability and a mind index for showing soil quality of dry and semi-dry lands. On the other hand, rangelands are one of the main dry ecosystems of carbon reservoirs. Knowing about carbon reservoir distribution and changes for detecting the controller mechanisms of the carbon world cycle and carbon stability is vital for managing ranges. Since field operations include soil sampling from different places and depths to measure the amount of soil carbon sequestration, it is very time-consuming and costly., On the other hand, different soil characteristics may be measured and available in many rangeland areas for other purposes using modeling and prediction under various inputs, including soil properties such as texture, acidity, electrical conductivity, etc. Researchers in the rangeland field can estimate and evaluate soil carbon. Novel prediction methods, including artificial intelligence, are of high interest in this field. The present research aims to study the ability of the adaptive neuro-fuzzy inference system (ANFIS) to predict the carbon sequestration (CS) of rangeland soil. Methods: The studied area in this research includes the rangelands of the southwestern slopes of Mount Damavand in the Lar watershed with an area of about 2000 hectares and an altitude between 2500 and 3460. With a statistical climate period of 36 years, it is a semi-humid to ultra-cold region where the average rainfall is 550 mm. Dominant plant species in the region are Onobrychis cornuta, Astragalus ochrodeucus, Astragalus microcephalus, Thymus pubescens, etc. Considering the Lar watershed region geographic conditions, four height groups relative to sea level, including height group 1 (2500-2700 m), group 2 (2700-2900 m), group 3 (2900-3100 m), and group 4 (3100-up m) were selected for sampling soil at different depths in this research. Thirteen random points were determined at all height groups, and three samples from each point were dug at depths 0-15 cm and 15-30 cm. In total, 312 soil samples were collected in the entire region and transferred to a soil science laboratory, where soil characteristics (texture, organic carbon, and bulk gravity) were measured as an average of three repetitions. These characteristics were used to calculate the amount of soil carbon deposition. After soil sampling and measuring the amount of carbon sequestration under the effect of soil depth and height in the sampling location at the Mount Damavand rangelands at the Lar watershed region, the regression and ANFIS prediction equations were developed and their accuracy was compared together for CS prediction plus introducing the more accurate approach. The root mean square error (RMSE) and correlation coefficient (R2) were applied to evaluate the regression and ANFIS models. The regression analysis was performed using the SPSS20 software. Excel software was used to draw descriptive charts. ANFIS modeling was created in MATLAB software and is based on the input/output dataset of a fuzzy inference system (FIS). This system is based on the combination of three components: membership functions of input and output variables (fuzzification), fuzzy rules (rule base), mechanism inference (a combination of rules with fuzzy input), output characteristics, and results of the system (de-fuzzification). Results: The results of the analysis of variance (ANOVA) revealed that only soil sampling depth significantly affected the soil CS, but the effect of sampling location and the interaction effect of depth and height were not significant. More amount of CS was obtained at a depth of 15-30 cm than at a depth of 0-15 cm, and the utmost amount of CS was measured in gang 4 of height (3119-3545 m) at both depths. The highest amount of CS (604656) belonged to gang 4 of height and a depth of 15-30 cm. In fact, the amount of soil CS increased at higher and lower altitudes while its amount decreased at medium altitudes. After gang 4 of height, the second most CS was recorded for the gang 1 of height. In the modeling part, the ANFIS model with higher accuracy (R2 = 0.4736) and lower error (RMSE = 0.0274) predicted the soil CS related to a regression model with lower accuracy (R2 = 0.4308) and higher error (RMSE = 0.069). This result indicates the higher ability of the ANFIS model than the regression model in creating a relationship between input and output and its proximity to the measured values. Conclusion: The increase in the correlation coefficient and the reduction of the mean error deviation in the ANFIS method compared to the linear multivariate regression show that the ANFIS method is more successful in estimating the amount of soil CS under the effects of various factors in the studied land use. The better performance of the ANFIS model than statistical regression methods can be found in its estimation and prediction capability for the nonlinear estimation with a small amount of data. This is despite the fact that the performance and accuracy of regression methods strongly depend on the sample size, and a small sample size can be a limiting factor in such statistical models. The adaptive neuro-fuzzy inference system (ANFIS) satisfied operation in predicting rangeland soil CS under the different sampling depths and heights. Furthermore, it will be used as an intelligent tool for predicting different parameters in studied ranges and rangeland science, such as above and underground biomass volume, distribution of rangeland plant species, and so on.