One particular strategy to render anticancer therapies efficient consists of converting the patient's own tumor cells into therapeutic vaccines, via the induction of immunogenic cell death (ICD). One of the hallmarks of ICD dwells in the active release of ATP by cells committed to undergo, but not yet having succumbed to, apoptosis. We observed that the knockdown of essential autophagy-related genes (ATG3, ATG5, ATG7 and BECN1) abolishes the pre-apoptotic secretion of ATP by several human and murine cancer cell lines undergoing ICD. Accordingly, autophagy-competent, but not autophagy-deficient, tumor cells treated with ICD inducers in vitro could induce a tumor-specific immune response in vivo. Cancer cell lines stably depleted of ATG5 or ATG7 normally generate tumors in vivo, and such autophagy-deficient neoplasms, upon systemic treatment with ICD inducers, exhibit the same levels of apoptosis (as monitored by nuclear shrinkage and caspase-3 activation) and necrosis (as determined by following the kinetics of HMGB1 release) as their autophagy-proficient counterparts. However, autophagy-incompetent cancers fail to release ATP, to recruit immune effectors into the tumor bed and to respond to chemotherapy in conditions in which autophagy-competent tumors do so. The intratumoral administration of ecto-ATPase inhibitors increases extracellular ATP concentrations, re-establishes the therapy-induced recruitment of dendritic cells and T cells into the tumor bed, and restores the chemotherapeutic response of autophagy-deficient cancers. Altogether, these results suggest that autophagy-incompetent tumor cells escape from chemotherapy-induced (and perhaps natural?) immunosurveillance because they are unable to release ATP.