Many reward learning theories posit that effort is aversive, indicating that reward system would be least responsive to difficult and effortful options. However, research in psychology suggests that effort itself can be rewarding. To test this possibility, we examined the relationship between task difficulty and reward network with 51 healthy participants using functional magnetic resonance imaging. Participants were assigned to three groups: Control group participants did not receive any rewards (i.e. they worked on the task out of intrinsic motivation), whereas reward group participants received monetary rewards based on their performance. In the gambling group, task success or failure was probabilistically determined (not by the performance of participants). In all groups, participants worked on a game-like task that had three difficulty levels: easy, moderate, and difficult. The behavioural results showed that the effects of difficulty level on self-reported motivation differed between the groups, F(2,96) = 22.72, p < .001. In the control group, participants were more motivated with increasing task difficulty, whereas the gambling group showed the opposite pattern. Furthermore, the motivation ratings in the reward group followed an inverted U shape, exhibiting the highest motivation for a moderately difficult task. The neuroimaging analyses mirrored these findings: a bilateral cluster in the ventral striatum showed similar patterns of activation as observed in behavioural analysis (p < .05, family-wise error corrected). Our results indicate that contrary to the current popular reward learning theories, effort has differential effects in the reward network depending on context and is not generally aversive.