Hypoxic fractions are measured by indirect techniques, which compare the response of tumors to large single doses of radiation given under normal aeration and artificial hypoxia. This paper reviews hypoxic fraction measurements and measurement techniques, giving particular attention to the biological, technical, and statistical aspects of the assays; the implicit assumptions underlying the analyses; and the dependence of the determinations on the assay conditions and the tumor and host characteristics. The three major hypoxic fraction assay techniques (paired survival curve, clamped tumor control, and clamped growth delay) share common biological assumptions. They require that the survival curves of naturally and artificially hypoxic cells have the same slope and intercept. They assume that the majority of the cells are either fully oxic or fully hypoxic. They assume that the methods used to induce artificial hypoxia leave no oxygenated regions and that tumor cells rendered artificially hypoxic are no less viable than cells in normally-aerated tumors. The universal validity of these assumptions is questionable. Each technique uses additional special assumptions and each may measure a different population of hypoxic cells. This paper reviews 92 hypoxic fraction determinations in 42 tumor systems. Radiobiologically hypoxic cells appear to be present in the majority of macroscopic solid rodent tumors. The hypoxic fraction was found to increase as the tumor size increased from microscopic to macroscopic; the dependence of hypoxic fraction on tumor size at macroscopic sizes was less clear. The site of tumor implantation, the use of anesthesia, and certain host characteristics may influence the hypoxic fraction. The hypoxic fraction generally did not depend on the tumor growth rate, transplantation history, or histology. These findings indicate that hypoxic cells are a common feature of solid tumors in rodents and provide no evidence that hypoxic cells should not be present in human tumors.