Goal-directed behavior requires the cognitive system to distinguish between relevant and irrelevant information. The authors show that task sets help to shield the system from irrelevant information. Participants had to respond to eight different colored word stimuli under different instruction conditions. They either had to learn the stimulus-response mappings (SR condition), to use one task set (1 TS condition) or to use two different task sets (2 TS condition). In the 2 TS and the SR conditions, participants showed response repetition effects (interaction of color repetition x response repetition), indicating that participants processed the color of the words. Importantly, the 1 TS condition did not show such an interaction. Overall, the results provide evidence for the shielding function of task sets. This benefit turns into costs in classical task switching paradigms. From this perspective, switch costs can be interpreted as the consequence of successful shielding on the previous task.