ABSTRACTTheoretical models propose that interoception plays a role in addictive disorder. However, this assumption has been mostly tested using the heartbeat counting task (HCT), which is known to be contaminated by estimation strategies. An adapted version of the HCT (in which respondents report only felt heartbeats) has been developed to reduce estimation biases. Here, we examined the validity of the classical and adapted HCT versions in samples presenting alcohol use disorders. We recruited a clinical sample of 48 patients with severe alcohol use disorder (SAUD), matched with 41 healthy controls (HC), and a subclinical sample of 32 binge drinkers (BD), matched with 30 HC. Participants performed the classical HCT, adapted HCT, and a time estimation task. We additionally assessed mental health variables theoretically related to interoception (alexithymia, anxiety, childhood trauma, depression and emotion regulation). In all groups, HCT scores were smaller in adapted than classical HCT. Patients with SAUD, but not BD, showed lower HCT scores than matched controls, independently of the task. We found no correlation between HCT scores and psychological constructs. Heartbeats reported during classical HCT correlated with seconds reported during time estimation task for SAUD and matched HC, suggesting the use of time estimation strategies to perform the task. The largely reduced HCT performance in the adapted version, the association between HCT performance and time estimation performance and the lack of theoretically expected associations between HCT scores and psychological variables extend doubts on the validity of these tasks for measuring interoceptive accuracy in problematic alcohol consumption.