Past studies have found that the in-group membership of an intelligent agent can improve users’ trust. We explored whether such an effect depends on system reliability levels and anthropomorphic appearance. We manipulated reliability levels (95%, 70%, 45%) and anthropomorphic appearance (human-like vs. computer-like appearance) of an intelligent decision support system in our study. The minimum group paradigm was adopted to manipulate the group membership of the intelligent system (in-group vs out-group). We measured trust by using both subjective rating and compliant behaviors toward the system recommendations. The results showed that the intelligent system with an in-group membership resulted in higher trust as compared to the system with an out-group membership. The magnitude of these effects did not differ across different reliability levels and anthropomorphic appearances. We discussed such findings in light of human-robot interaction theories and potential implications for designing trustworthy decision support system.