The mismatch negativity (MMN) is conceptualized as a confidence-weighted error signal elicited when a deviation violates the predicted next-state based on regularity. The mechanisms underpinning its generation remain contentious. Smaller MMN response is a robust finding in schizophrenia and reduced amplitude may implicate impairment in prediction-error signalling. An enriched understanding of factors that influence MMN size in healthy people is a prerequisite for translating the relevance of reduced MMN in schizophrenia. This paper features two studies designed to explore factors that impact MMN in healthy individuals. Study 1 confirms that MMN amplitude does not faithfully reflect transition statistics and is susceptible to order-driven bias. In study 2, we demonstrate that an order-driven bias remains despite repeated encounters with sound sequences. These data demonstrate that factors that impact on MMN size in non-clinical groups are not fully understood and that some mechanisms driving relevance filtering are likely influenced by ‘top–down’ expectations.