Abstract Mutation of the phosphatidylinositol N-acetylglucosaminyltransferase subunit A (PIG-A) gene in hematopoietic stem cells (HSCs) results in the loss of glycosylphosphatidylinositol-anchored proteins (GPI-APs) on HSCs, but minimally affects their development, and thus can be used as a clonal maker of HSCs. We analyzed GPI-APs expression on six major lineage cells in a total of 574 patients with bone marrow (BM) failure in which microenvironment itself is thought to be unaffected, including aplastic anemia (AA) or myelodysplastic syndrome (MDS). GPI-APs-deficient (GPI-APs−) cells were detected in 250 patients. Whereas the GPI-APs− cells were seen in all six lineages in a majority of patients who had higher proportion ([dbmtequ]3%) of GPI-APs− cells, they were detected in only limited lineages in 92.9% of cases in the lower proportion (<3%) group. In all 250 cases, the same lineages of GPI-APs− cells were detected even after 6–18-month intervals, indicating that the GPI-APs− cells reflect hematopoiesis maintained by a self-renewing HSC in most of cases. The frequency of clones with limited lineages seen in mild cases of AA was similar to that in severe cases, and clones with limited lineages were seen even in two health volunteer cases. These results strongly suggest most individual HSCs produce only restricted lineages even in a steady state. While this restriction could reflect heterogeneity in the developmental potential of HSCs, we propose an alternative model in which the BM microenvironment is mosaic in supporting commitment of progenitors toward distinct lineages. Our computer simulation based on this model successfully recapitulated the observed clinical data.