It has been demonstrated that the Simon effect may be increased or reversed due to proportion congruency manipulation, suggesting that learned spatial irrelevant stimulus-response (S-R) associations are used to guide responses. In this study, we tested the hypothesis that learning spatial irrelevant S-R associations by rewards may show a similar modulatory effect on the Simon effect. In two experiments with the Simon task, we manipulated the contingency of stimulus-response-reward between groups. Experiment 1 showed that the Simon effect in both reaction times and error rates increased if potential performance-contingent rewards always followed congruent trials but decreased and was even reversed if rewards always followed incongruent trials. These suggest that participants used reward-strengthened spatially compatible and incompatible irrelevant S-R associations respectively to predict responses. Experiment 2 showed that the data pattern of the increase and reversal of the Simon effect showed in both rewarded and nonrewarded colors, suggesting that reward-strengthened spatial irrelevant S-R associations were used to guide responses even when there were no potential rewards. Together, these results resemble the proportion congruency effect with the Simon task, suggesting that there could be stronger conflict and attentional control when the correct response is different from the response activated by reward-strengthened irrelevant S-R associations. This suggests that reinforcement learning of irrelevant S-R associations can modulate cognitive control.