In malignant gliomas, the characteristically heterogeneous features and frequent diffuse spread within the brain have raised the question of whether malignant gliomas arise monoclonally from a single precursor cell or polyclonally from multiple transformed cells forming confluent clones. Although monoclonality has been shown in surgically resected tissues, these may not include the full spectrum of patterns seen on autopsy material. Little is known about the clonality of low-grade gliomas from which malignant gliomas may sometimes arise. We sought to investigate the clonality of low-grade and malignant gliomas by using and comparing surgical and autopsy material with a Polymerase chain reaction (PCR)-based assay for nonrandom X chromosome inactivation. For that, purpose, archival surgical and autopsy material from 15 female patients (group A) (age 4 to 73 years; median, 45) with malignant gliomas (12 glioblastomas, one gliosarcoma, one anaplastic oligoastrocytoma, one gliomatosis cerebri), surgical material only from 21 female patients (group S) (age 6 to 78 years; median, 60) with low-grade and malignant gliomas (four low-grade astrocytomas, three oligoastrocytomas, two anaplastic astrocytomas, one gemistocytic astrocytoma, four oligodendrogliomas, seven glioblastomas) were analyzed. In group A, representative areas (mean = 5/patient; median = 7) were microdissected from tissue sections and assayed by PCR amplification of a highly polymorphic microsatellite marker locus of the human androgen receptor gene (HUMARA) in the presence of alpha32P with and without predigestion with a methylation-sensitive restriction enzyme (HhaI). Products were resolved by denaturing gel electrophoresis and autoradiographed. In group S, selected tumor areas were used for the assay. Each patient's normal brain tissue was used for control. The band intensity of alleles were measured by densitometric scanning. In group A, 13 of 15 cases were informative (heterozygous). The same pattern of nonrandom X chromosome inactivation was present in all areas of solid dense and moderate tumor infiltration in eight including all components of the gliosarcoma. Two of eight also showed focal loss of heterozygosity (LOH). One of 13 presented global LOH. Two of 13 showed microsatellite instability, one of which in a patient with Turcot syndrome, the other in gliomatosis cerebri. Opposite skewing patterns were seen in distant areas of gliomatosis cerebri consistent with oligoclonal derivation. Clonality remained indeterminate in one glioblastoma and in the anaplastic oligoastrocytoma because of skewed lyonization in the normal control. In group S, 19 of 21 cases were informative. Fifteen of 19 were monoclonal (four low-grade astrocytomas, one anaplastic astrocytoma, one gemistocytic astrocytoma, two oligodendrogliomas, one oligoastrocytoma, six glioblastomas). Four of 19 were indeterminate. We conclude that (1) Low-grade and malignant gliomas are usually monoclonal tumors, and extensively infiltrating tumors must result from migration of tumor cells (2) Gliomatosis cerebri may initiate as an oligoclonal process or result from collision gliomas (3) Biphasic gliomas likely arise from a single precursor cell. (4) LOH at the HUMARA locus is probably related to partial or complete deletion of an X-chromosome, which occurs in malignant gliomas during clonal evolution.