Abstract: An accurate estimation of earthwork can greatly contribute to characterizing the deformation during land leveling and land reclamation of mining areas, even to monitor the soil erosion under various surface coverage. However, it cannot be well understood to estimate the earthwork using Terrestrial Laser Scanning (TLS). In this study, the TLS was selected to evaluate the earthwork under different surface conditions (complex terrain, different soil gravel content, and vegetation coverage). Seven experimental plots were set up in the Red Soil Ecological Experimental Station in Yingtan City, Jiangxi Province of China. The length and width of Plots 1-6 were 10.0 m and 2.5 m with a slope of 8, while Plot 7 was in the complex terrain with an average slope of 20. The vegetation coverage was varied in the Plots 1~6, where no vegetation (Plot 2), sparse grass only (Plot 1, 0.57%), masson pine only (Plot 4 and Plot 6, 21.92% and 28.47%), sparse grass, and masson pine (Plot 5, 26.12%). Meanwhile, the soil gravel was input to Plot 3 and Plot 4. The soil was excavated from Plot 7, and then filled to Plots 1-6. In each plot, the TLS datasets were acquired before and after earth filling/excavation. A digital elevation model (DEM) was then obtained using the moving curved fitting filtering and inverse distance weighted interpolation. The height change of DEM and the bulk density were selected to estimate the mass of earth filling/excavation, further to be validated by the measured mass of earthwork. In addition, the performance of estimated earthwork was also evaluated using the number, the location of the TLS station, and the neighboring point distance of TLS data. The results showed that: 1) The average absolute relative errors (ARE) of estimated earthwork for the different plots were ranked as the Plot 7 (complex terrain, 18.17%) > Plot 3 (soil gravel, 10.30%) > Plot 4 (masson pine and soil gravel, 9.53%)> Plot 1 (sparse grass, 7.48%) > Plot 6 (masson pine forest, 2.26%) > Plot 2 (bare surface, 2.13%) > Plot 5 (masson pine and sparse grass, 1.69%). There were the underestimated and overestimated earthwork for the Plots 1-6, and Plot 7, respectively; 2) When the vegetation coverage was less than 1% and ranged from 20% to 30%, the number of stations were reduced to two and three, respectively, where the estimation accuracy was ≥99% of the original. The estimation error of earthwork was lower using the TLS scan in the downslope than that in the upslope when the height of the instrument was not higher than 1.5 m and the number of TLS scan was one. Once the number of TLS scan reached two, the estimated accuracy was higher than before, particularly when the TLS scan was set in the diagonal direction; 3) The estimation error increased significantly, with the increasing average neighboring point distance. The optimal distances of average neighboring points were 1-8 cm for the Plots with almost nothing coverage near the ground, and the remaining four Plots covered by grass or soil gravel near the ground were 1-4 cm. This finding can provide potential guidance to explore the spatiotemporal variations of earthwork in the centimeter-level using the TLS in the field.