In a reaction task responding has often proved faster when the levels of the independent variable are presented isolated from each other, in pure blocks than when they are presented randomly intermixed, in mixed blocks. This difference in response time is denoted here as mixing costs. This paper presents a theoretical review of mixing costs with special emphasis on their origin. In Section 1, two views on the origin of mixing costs are delineated, which are subsequently elaborated in Sections 2 and 3. The strategic view asserts that subjects are less well prepared in mixed blocks than in pure blocks, due to greater uncertainty about the level of the independent variable to be presented on the forthcoming trial. The stimulus-driven view attributes mixing costs to more pronounced trial-to-trial adaptation of processing in mixed blocks than in pure blocks, due to greater intertial variability. Section 4 reviews mixing costs deriving from various areas of human performance, and concludes that the dominant strategic view in the literature is not warranted, and that stimulus-driven factors have been underestimated.