The observed seasonal cycles in atmospheric potential oxygen (APO) at a range of mid-to high-latitude surface monitoring sites are compared to those inferred from the output of six Earth system models (ESMs) participating in the fifth phase of the Coupled Model Intercompar-ison Project phase 5 (CMIP5). The simulated air-sea O 2 fluxes are translated into APO seasonal cycles using a matrix method that takes into account atmospheric transport model (ATM) uncertainty among 13 different ATMs. Three of the ocean biogeochemistry models tested are able to reproduce the observed APO cycles at most sites, to within the large TransCom3-era ATM uncertainty used here, while the other three generally are not. Net primary production (NPP) and net community production (NCP), as estimated from satellite ocean color data, provide additional constraints, albeit more with respect to the seasonal phasing of ocean model productivity than overall magnitude. The present analysis suggests that, of the tested ocean biogeochemistry models, the community ecosystem model (CESM) and the Geophysical Fluid Dynamics Laboratory (GFDL) ESM2M are best able to capture the observed APO seasonal cycle at both northern and southern hemispheric sites. In most models, discrepancies with observed APO can be attributed to the underestimation of NPP, deep ventilation or both in the northern oceans.