Purpose The Hospital Anxiety and Depression Scale (HADS) is a self-report questionnaire designed to screen anxious and depressive states in patients in non-psychiatric settings. In spite of its large use, no agreement exists in literature on HADS accuracy in case finding. The present research addresses the issue of HADS accuracy in cancer patients, comparing its two subscales (HADS-A and HADS-D) against tools not in use in psychiatry, which are able to detect prolonged negative emotional states. Methods 2121 consecutive adult cancer inpatients were administered the HADS together with the State Anxiety subscale of State-Trait Anxiety Inventory and the Center for Epidemiologic Studies Scale on Depression. Receiver operating characteristic (ROC) curves were computed to identify a cut-off for anxious and depressive states in cancer patients. All indicators were computed together with their corresponding 95% confidence interval (95% CI). Results Data of 1628 and 1035 participants were used to assess the accuracy in case finding of HADS-A and HADS-D, respectively. According to the ROC analysis, the optimal cut-off was > 9 units for the HADS-A and > 7 units for the HADS-D. The area under the ROC curve was 0.90 for HADS-A (95% CI 0.88–0.91) and 0.84 for HADS-D (95% CI 0.81–0.86). Conclusions This study suggested that risk scores of anxious and depressive states above specific HADS cut-offs are useful in identifying anxious and depressive states in cancer patients, and they may thus be applicable in clinical practice.