Development of an appropriate scheme to accurately sample root biomass is essential for accurate estimation of biomass and carbon budget of grassland. This study evaluated measurement accuracy of the monolith and core methods with different sample sizes and positions in a temperate grassland ecosystem in Inner Mongolia, China. The results indicated that the small core method (3.8-cm-diameter) significantly underestimated total root biomass compared with the large core method (10-cm-diameter), small monolith method (0.25 m(2)) and large monolith method (1 m2). Total root biomass estimated from the small core method was about 52% less than that from the large monolith method (1 m(2)). At 95% confidence interval, 10% relative precision could be obtained with five small monoliths, 15 large cores and 65 small cores. The coefficient of variation (CV) for total root biomass decreased logarithmically with increasing sample size for both the monolith and core methods. Compared with the stratified random sampling, core sampling with different fixed positions could not provide reliable estimate of total root biomass. Washing damage and soil lost during extraction might be the major factors controlling the measurement accuracy of total root biomass by core method with small sample size. (C) 2009 Elsevier B.V. All rights reserved.