Abstract: The climatologies of dissolved oxygen concentration in the ocean simulated by nine Earth system models (ESMs) from the historical emission driven experiment of CMIP5 (Phase 5 of the Climate Model Inter-comparison Project) are quantitatively evaluated by comparing the simulated oxygen to the WOA09 observation based on common statistical metrics. At the sea surface, distribution of dissolved oxygen is well simulated by all nine ESMs due to well-simulated sea surface temperature (SST), with both globally-averaged error and root mean square error (RMSE) close to zero, and both correlation coefficients and normalized standard deviation close to 1. However, the model performance differs from each other at the intermediate depth and deep ocean where important water masses exist. At the depth of 500 to 1 000 m where the oxygen minimum zones (OMZs) exist, all ESMs show a maximum of globally-averaged error and RMSE, and a minimum of the spatial correlation coefficient. In the ocean interior, the reason for model biases is complicated, and both the meridional overturning circulation (MOC) and the particulate organic carbon flux contribute to the biases of dissolved oxygen distribution. Analysis results show the physical bias contributes more. Simulation bias of important water masses such as North Atlantic Deep Water (NADW), Antarctic Bottom Water (AABW) and North Pacific Intermediate Water (NPIW) indicated by distributions of MOCs greatly affects the distributions of oxygen in north Atlantic, Southern Ocean and north Pacific, respectively. Although the model simulations of oxygen differ greatly from each other in the ocean interior, the multi-model mean shows a better agreement with the observation.