Fifteen permanent cell lines derived from human gliomas which are individually distinct by immunologic and biochemical criteria were evaluated to determine if morphologic or cell biologic parameters distinguished the 4 lines which were tumorigenic in athymic nude mice. By subjective morphologic appraisal, the 4 tumorigenic lines were considered "malignant" or "borderline," but 4 of the non-tumorigenic lines were also classified in this way. By objective criteria, these 15 lines varied markedly in percentage of piled-up cells, chromatin pattern, pleomorphism, nuclear to cytoplasmic ratio, number of bizarre multinucleate giant cells, presence of abnormal mitotic figures, percentage of colony formation in soft agar, saturation density, population doubling time, and absolute plating efficiency. Among these criteria, percentage of colony formation in soft agar had the highest correlation coefficiency with tumorigenicity, and when this parameter was held constant the only additional characteristic which correlated significantly (p less than .05) was the number of bizarre multinucleate giant cells. When the 11 non-tumorigenic lines were ranked by these 2 criteria, 1 non-tumorigenic line (U-251 MGsp) had greater than .95 predicted probability of tumorigenicity. Although further tumorigenicity testing may increase the number of tumorigenic lines, the lines with few "malignant" characteristics may correspond to the population resembling cells of low grade astrocytomas seen within glioblastomas. The histologic pleomorphism of human gliomas is reflected in their morphologic and cell biologic diversity in culture.