The response set effect refers to the finding that an irrelevant incongruent colour-word produces greater interference when it is one of the response options (referred to as a response set trial), compared to when it is not (a non-response set trial). Despite being a key effect for models of selective attention, the magnitude of the effect varies considerably across studies. We report two within-subjects experiments that tested the hypothesis that presentation format modulates the magnitude of the response set effect. Trial types (e.g. response set, non-response set, neutral) were either presented in separate blocks (pure) or in blocks containing trials from all conditions presented randomly (mixed). In the first experiment we show that the response set effect is substantially reduced in the mixed block context as a result of a decrease in RTs to response set trials. By demonstrating the modulation of the response set effect under conditions of trial type mixing we present evidence that is difficult for models of the effect based on strategic, top-down biasing of attention to explain. In a second experiment we tested a stimulus-driven account of the response set effect by manipulating the number of colour-words that make up the non-response set of distractors. The results show that the greater the number of non-response set colour concepts, the smaller the response set effect. Alternative accounts of the data and its implications for research debating the automaticity of reading are discussed.