We hypothesize that impaired behavioral flexibility might underlie an array of symptoms associated with autism spectrum disorder (ASD), especially repetitive and restricted behavior. Reversal learning is a simple test for behavioral flexibility, and previous studies have shown that ASD patients and BTBR T+tf/j (ASD model) mice are impaired in probabilistic reversal learning. As an attempt to develop an efficient behavioral paradigm to test behavioral flexibility in ASD model mice, we tested Shank2 knockout (Shank2-KO) mice in a probabilistic reversal learning task under a head-fixed condition. Two different odor cues were presented to a head-fixed mouse and paired with a reward (water) or a punishment (air puff) each with 75% probability. Head fixation prevented Shank2-KO mice from emitting abnormal hyper-excitable behaviors, such as excessive grooming and jumping, but Shank2-KO mice showed normal licking behavior in response to water delivery. Both Shank2-KO and wild type mice showed higher anticipatory licking responses to the reward-predicting than punishment-predicting cues, indicating intact learning of cue-outcome contingency in both animal groups. However, upon the reversal of cue-outcome contingency, it took much longer for Shank2-KO than wild type mice to show higher anticipatory licking responses to the new reward-predicting cue. These results suggest our probabilistic reverse learning task as a potentially useful behavioral paradigm for testing behavioral flexibility in ASD model mice.