Because of the insensitivity of the Ham test, paroxysmal nocturnal haemoglobinuria (PNH) has been inaccurately viewed as a late clonal complication of aplastic anaemia (AA). To clarify the relationship between PNH and marrow failure, we tested for the presence of glycosylphosphatidyl-anchored protein-deficient (GPI-AP) granulocytes in large cohorts of patients with AA, myelodysplasia (MDS), and pure haemolytic PNH. A PNH clone was detected in 32% of new AA patients and 18% of MDS patients. In serial studies, this proportion did not change up to 15 years after diagnosis, suggesting that expansion of aberrant cells is an early event (i.e. prior to initial presentation). For all patients with a PNH clone, on average 14% of PNH granulocytes were found on presentation and 37% at 10 years. Patients with PNH but without cytopenia showed higher percentages of GPI-AP-deficient cells than did those with the AA/PNH syndrome. After immunosuppression, there was no change in the contribution of PNH clone to blood production, arguing against the "immune-escape" theory in PNH. Clinically, a high proportion of GPI-AP-deficient cells correlated with marrow hypercellularity. GPI-AP-deficient cells were similarly present in patients with and without karyotypic abnormalities. Our results indicate that the GPI-AP-deficient clones show quantitative and kinetic differences between classic haemolytic PNH and PNH with marrow failure, in which the evolution rate is low later in the course of the disease.