Past work has documented that the detrimental effect of dual-tasking on learning a repeating sequence of stimulus positions and responses in the Serial Reaction Time Task (SRTT) is due to lack of separation of representations: Participants encode information about the stimuli and responses of the two tasks in joint memory episodes. Therefore, when a task with random stimulus order is paired with the SRTT, encoded compounds contain randomness as well. In the current work we explore whether across-task integration in dual-tasking can be avoided by learning a sequence beforehand in single-tasking. Participants (N=39) reacted to a repeating fixed sequence of six stimulus positions throughout the experiment. In Blocks 1 to 3 (single-tasking) they learned the sequence. This was shown by the RT advantage compared to a control group with random stimulus order (N=41). In Blocks 4 to 6 (dual-tasking) a randomly located stimulus was additionally present in each trial. RTs showed that sequence knowledge could now no longer be accessed. Furthermore, across-task integration was evident no matter whether or not a fixed sequence had been learned before dual-tasking: RTs for trials in which the random stimulus by chance was the same as 6 trials ago (last loop of the sequence) were shorter compared to alternation trials.