Endostatin, a 20-kDa carboxy-terminal fragment of collagen XVIII, is the leading member of a class of physiologic inhibitors of angiogenesis with potent antitumor activity. Repeated subcutaneous administration of recombinant endostatin in mice led to permanent regression of established tumors to a microscopic dormant state and prompted the initiation of human clinical trials. However, a discrepancy remained unresolved: sustained tumor regression has only been observed with a non-soluble, precipitated form of recombinant endostatin produced in bacteria. To shed light on this question and establish a model of systemic anti-angiogenic gene therapy of cancer that may surmount obstacles in protein production and delivery, we transduced murine hematopoietic stem cells with a retrovirus encoding a secretable form of endostatin. Despite continuous, high-level secretion of endostatin in the vasculature of all transplanted mice, we detected neither inhibition of in vivo neoangiogenesis nor antitumor activity. Resolution of this paradox may come from human trials of endostatin now underway.