American bison (Bison bison L.) have recovered from the brink of extinction over the past century. Bison reintroduction creates multiple environmental benefits, but impacts on greenhouse gas emissions are poorly understood. Bison are thought to have produced some 2 Tg yr−1 of the estimated 9–15 Tg yr−1 of pre-industrial enteric methane emissions, but few measurements have been made due to their mobile grazing habits and safety issues associated with measuring non-domesticated animals. Here, we measure methane and carbon dioxide fluxes from a bison herd on an enclosed pasture during daytime periods in winter using eddy covariance. Methane emissions from the study area were negligible in the absence of bison (mean ± standard deviation = −0.0009 ± 0.008 µmol m−2 s−1) and were significantly greater than zero, 0.048 ± 0.082 µmol m−2 s−1, with a positively skewed distribution, when bison were present. We coupled bison location estimates from automated camera images with two independent flux footprint models to calculate a mean per-animal methane efflux of 58.5 µmol s−1 per bison, similar to eddy covariance measurements of methane efflux from a cattle feedlot during winter. When we sum the observations over time with conservative uncertainty estimates we arrive at 81 g CH4 per bison d−1 with 95 % confidence intervals between 54 and 109 g CH4 per bison d−1. Uncertainty was dominated by bison location estimates (46 % of the total uncertainty), then the flux footprint model (33 %) and the eddy covariance measurements (21 %), suggesting that making higher-resolution animal location estimates is a logical starting point for decreasing total uncertainty. Annual measurements are ultimately necessary to determine the full greenhouse gas burden of bison grazing systems. Our observations highlight the need to compare greenhouse gas emissions from different ruminant grazing systems and demonstrate the potential for using eddy covariance to measure methane efflux from non-domesticated animals.