Abstract: Taking the coniferous forests in Shangri-La City as the object of research and determine the 95% reliability index and 85% sampling accuracy, a total of 10 first-order sampling units specifications of 1 km×1 km, 2 km×2 km, 3 km×3 km, 4 km×4 km, 5 km×5 km, 6 km×6 km, 7 km×7 km, 8 km×8 km, 9 km×9 km and 10 km×10 km were selected in the second-order sampling(SOS) method, then an optimal design of SOS was carried out based on the variance difference between first-order units elements of SOS. Finally, the sampling error and sampling accuracy of both sampling methods were compared and analyzed. The results showed that among the 10 kinds of SOS, the sampling error of the 6 km×6 km has the lowest error and highest accuracy, and the sampling error is 9.12%, the sampling accuracy is 90.88%. After the optimization, the sampling accuracy of SOS are improved and sampling error decreased significantly. Especially, the sampling error of the 6 km×6 km sampling unit is 6.39% and a sampling accuracy of 93.61%. Therefore, the optimized SOS has the better overall accuracy and practicability, which could be used to forest biomass investigation with the accuracy guarantee.