The size of response-repetition (RR) costs, which are usually observed on task-switch trials, strongly varies between conditions with univalent and bivalent stimuli. To test whether top-down or bottom-up processes can account for this effect, we assessed in Experiment 1 baselines for univalent and bivalent stimulus conditions (i.e., for stimuli that are associated with either 1 or 2 tasks). Experiment 2 examined whether the proportion of these stimulus types affects RR costs. As the size of RR costs was independent of proportion, a top-down explanation could be excluded. However, there was an increase in RR costs if the current stimulus induced a response conflict. To account for this effect, we proposed an amplification of response conflict account. It assumes that the basic mechanism that leads to RR costs amplifies response conflict, which, in turn, increases RR costs. Experiment 3 confirmed this bottom-up explanation by showing that the increase in RR costs varies with previous-trial congruency, which is known to affect RR costs. Experiment 4 showed that the increase can also be found with univalent stimuli that induce response conflict. Altogether, the results are in line with a response inhibition account of RR costs. Implications for alternative accounts are also discussed. (Contains 5 figures, 3 tables and 6 footnotes.)