Human activities in modifying land use and land cover increasingly put pressure to many regulatory ecosystem services, one of which is carbon sequestration. If forests, the area with the most vegetation cover are decrease, the amount of carbon sequestered will decrease significantly. Currently, agroforestry systems or Talun (in West Java) in Sumedang was eleven times larger than secondary forest. Carbon stocks in this agricultural area need to be estimated so that their carbon sequestration capacity can be known in order to improve the quality of regulatory ecosystem services. NDVI value of the Landsat 8 OLI was obtained by conducting raster calculation in ArcMap. Field inventory was conducted by measuring stem DBH and height of all vegetation stands in 31 plots measuring 30 x 30 m, a similar plot size to the resolution of the Landsat imagery. Biomass of vegetation stands was calculated using allometric equations and then converted into the carbon content of the biomass. In order to analyze the correlation of NDVI and carbon inventory data, a Pearson product-moment correlation analysis in the form of simple linear regression, non-linear exponential, and polynomial order 2 and order 3 model were carried out. Standard error of estimate (SEE) was performed to identify the best equation to model the aboveground carbon stocks in the area. The results show that the four regression models give a positive correlation between NDVI and carbon stocks. The strongest category was the polynomial order 2 and order 3 regression model with 0.795 coefficient of determination. Yet, the linear simple regression model obtained the highest accuracy with estimated error 0.445 tons/pixel. The estimated carbon stock obtained by linear regression model was 16150.40 tons with an average of 104.95 tons/ha. Visually, according to carbon distribution map, the carbon content of vegetation stands in the mixed garden of Rancakalong was mainly distributed in the north of Rancakalong District which is located in Cibungur Village.