In a prospective long-term study on the incidence of paroxysmal nocturnal hemoglobinuria (PNH), 115 consecutive patients with severe aplastic anemia (SAA), 97 treated with antilymphocyte globulin (ALG) and 18 with bone marrow transplantation (BMT), were observed over a period of 4–18 years and tested for the presence of complement-sensitive hematopoietic precursor cells with the bone marrow (BM) sucrose test. Sixteen (14%) of the ALG-treated patients developed clinical signs of PNH between 0.5 and 8 years after treatment. Complement-sensitive BM precursors were found in 89% of the SAA patients at some time during their disease, but in none of 18 normal donors. At diagnosis, their proportion was significantly higher in patients who later developed PNH than in patients who later achieved disease-free complete remission (CR). After ALG, the abnormal population was found in both groups, but it was gradually replaced by normal precursors in remission patients. After BMT, the complement-sensitive population decreased to very low numbers in patients with a stable graft, but increased again in 3 patients upon graft rejection. Mimicking the PNH defect by enzymatic removal of glycosyl-phosphatidylinositol (GPI)-linked proteins from CD34+ cells resulted in their complement sensitivity, suggesting that the BM sucrose test identifies precursor cells carrying the PNH defect. In 66 patients, white blood cells (WBC) in peripheral blood (PB) were examined for GPI-deficient populations by flow cytometry (FACS). Ten patients with signs of clinical or laboratory PNH had over 25% complement-sensitive precursor cells in the BM and a GPI-deficient WBC population in the PB. Of 56 SAA patients without PNH, 8 had an abnormal population detectable with both tests, 26 only with the BM sucrose test, 4 only with PB FACS analysis, and in 18, no abnormal cells were detected with either test. In search for parameters which might explain why in some patients the abnormal population expands, while it regresses or disappears in others, we tested the release of IL-2 as a parameter of immune competence. At diagnosis, IL-2 release was approximately 50% of normal in patients who later developed PNH, while it was double the normal value in patients who later achieved CR. We conclude that the majority of SAA patients transiently harbor complement-sensitive precursor cells in the BM. Patients with more than 25% abnormal BM precursors and low endogenous IL-2 release are at risk of progression to clinical PNH.