Most climate models underestimate the frequency of Atlantic blocking. Horizontal resolution is often cited as the main culprit due to poorly resolved small-scale variability whose upscale effects help to maintain blocks. However, recent studies show that blocking errors are also largely attributable to the large scale climatological bias of the model. Furthermore, modest resolution models can contain enough variability to generate greatly improved blocking frequency if they are corrected to account for time-mean bias. Here we show greatly improved simulations of Atlantic winter blocking frequency in a coupled ocean–atmosphere climate model. A reduction of the mean bias, due to an improved simulation of the Atlantic Ocean, is a key element of the improvement.