In three experiments, the authors supported the hypothesis that parallel response activation seen in dual-task performance results from holding Task 2 rules in working memory (WM) while performing Task 1. To this end, the authors used the backward compatibility effect (BCE; quicker primary responses when the Task 2 response is compatible with codes of Task 1) as a marker for parallel response activation and manipulated WM load. Increasing the number of primary task rules from two to four did not modulate BCE, replicating Hommel and Eglau (2002), but a higher load condition, involving six primary task rules, reduced the BCE to nonsignificant levels. Experiment 3 further showed that WM is loaded by rules associating abstract stimulus categories to responses, and not by rules that associate individual stimuli to responses (S-R rules).