Perceptual and working memory tasks involving visual orientation are often characterized by two behavioral effects: the oblique effect, where precision for cardinal orientations is higher than for oblique orientations, and the repulsion bias, where orientation reports are biased away from cardinal orientations. It has been proposed that both effects can be explained by a Bayesian framework based on efficient coding. Despite this common mechanism, each effect in isolation would predict distinct neural representations – while the oblique effect would lead to stronger and more distinct neural signals for cardinals compared to obliques, the repulsion bias would split orientation space along the cardinal axes into representations clustered around either oblique. Here, we tested these predictions by reanalysing EEG data from several visual working memory experiments using orientation stimuli. We used representational similarity analysis to explicitly test models of the oblique effect and repulsion bias during orientation perception and working memory maintenance. During perception, there was a larger representational separation between the two cardinal axes compared to the two oblique axes, as predicted by the oblique model, but no evidence for repulsion bias. In contrast, working memory representations were more in line with the repulsion bias model. These results provide neural evidence for both the oblique effect and repulsion bias during different stages of visual working memory tasks. While the oblique effect is likely driven by a visual system optimized to perceive and distinguish cardinal orientations above all other orientations, the repulsion bias appears to be the result of a coarse post-perceptual clustering in lower-resolution parts of orientation space during working memory maintenance.