Antarctic coastal ocean warming is a major driver of uncertainty in future sea level rise, by driving ice sheet retreat via the submarine melting of floating ice shelves.  We explore how and why predictions of ocean warming differ across climate models in order to target model improvements and to characterize uncertainty in future predictions.  We find that climate models tend to be biased by about 1/2 degree Celsius in certain sectors of the Antarctic coastline, with substantial inter-model differences in the Amundsen Sea sector that are likely due to differences in how the models simulate changes to the global ocean circulation.  The Ross and Weddell seas show similarities in their warming that are probably less related to global climate change and more related to local physical processes that they share in common, such as sea ice behavior and regional wind patterns.  A few models show particularly high rates of Ross Sea warming, but this may be an artifact of their coarse resolution and biases in simulating the regional circulation.  These results point the way toward targeting processes that should be improved in climate models and take steps toward quantifying the uncertainty Antarctic sea level rise resulting from multi-model differences in predicted ocean warming.