Dual systems accounts have dominated research into human category learning. Despite this, there has been limited investigation into how people switch between proposed systems. We report an experiment which modified a popular probabilistic categorisation task (the Weather Prediction Task) in an attempt to demonstrate conditions under which such a switch could occur. The results suggested that increasing working memory load impaired categorisation performance and reduced the flexibility with which participants applied their knowledge. Importantly, no clear evidence of a switch between learning systems was found despite using a design intended to favour a shift from declarative to procedural learning. These findings pose questions for the expectation that the nature of a category structure will determine which system is engaged.