Based on field data collected at nine roundabouts in Gold Coast, Australia, this paper evaluates the performance of the capacity estimation for single-lane roundabouts using analytical models (including the Highway Capacity Manual (HCM) 2000 model, the German Highway Capacity Manual (GHCM) model, the Signalized & unsignalized Intersection Design and Research Aid (SIDRA) model and a new roundabout capacity (NRC) model) and an empirical model (the HCM 2010 model). We firstly calibrate critical gaps, follow-up times and conflicting flows. Compared with the capacities measured in the field, we carry out a study to analyse the accuracy of the HCM, GHCM, SIDRA and NRC models. The resultsshow the following: (1) The five models underestimate capacity but the NRC model produces a smaller range of relative error (-1.07% to -5.74%) and root-mean-square deviation (47.68) than the HCM 2010 (-4.62% to -16.14% and 105.00, respectively), HCM 2000 (-5.76% to -17.21% and 115.29, respectively), GHCM (-8.95% to -21.26% and 146.82, respectively) and SIDRA (-5.15% to -17.51% and 113.48, respectively) models; accordingly, the NRC outperforms the other four models for the surveyed roundabouts, (2) The five models estimate a similar capacity in low and medium traffic volumes, and exiting vehicles play an important role in high traffic volumes, and (3) At high traffic condition, exiting vehicles