Current models of visual working memory (VWM) assume that comparing memory with the environment obligato-rily involves a spatial comparison process. Can changing task demands determine whether a spatial or non-spatial comparisonprocesses is employed? Study displays of three colored shapes were presented, followed by test displays of three colouredshapes. Participants decided whether a feature changed between displays. Task-irrelevant changes to the probed item’s lo-cations or feature bindings reduced memory performance, suggesting that participants employed spatially guided comparisonprocess. This finding occurred irrespective of whether participants decided about the whole display, or only a single cueditem within the display. When task-irrelevant feature changes occurred amongst uncued items, performance was unaffected byirrelevant changes in location or feature bindings. These results suggest that participants can flexibly shift comparison strat-egy in response to changing task demands. These findings have implications for models of VWM, which assume obligatorylocation-based comparisons in VWM.