Animals and humans have multiple reinforcement learning (RL) systems, especially a hard-wired Pavlovian system that learns state-outcome associations and promotes an approach to rewards and avoidance of punishment; and an instrumental system that learns values based on state-action-outcome associations. Conflict between the two systems sometimes leads to suboptimal decisions including pathological behaviors but factors that affect the conflict remain largely unclear. Meanwhile, previous findings suggest that taxing cognitive resources alters several decision-making processes and a balance between multiple decision-making systems: under working memory (WM) load, 1) the reliance on computationally cheaper model-free RL increased compared to model-based RL; 2) learning rates in model-free RL decreased; 3) choice consistency decreased. However, the role of WM in the Pavlovian-instrumental conflict remains unknown. Thus, we conducted a functional magnetic resonance imaging (fMRI) study (N=49) in which participants completed a model-free RL task with Pavlovian-instrumental conflict (the orthogonalized go/no-go (GNG) task), and WM load was manipulated with dual-task conditions. The behavioral and computational modeling analysis showed that WM load did not affect the Pavlovian bias, but instead model-free RL by decreasing learning rate and increasing random choice. fMRI analysis showed that striatal reward prediction error signaling increased under WM load. Moreover, under WM load, the striatum showed weakened connectivity with the ventromedial and dorsolateral prefrontal cortex when computing reward expectation. These results suggest that limited cognitive resources do not directly affect the competition between the instrumental and hard-wired Pavlovian systems, but influence model-free RL and action selection through the weakened cooperation for learning between WM and RL.