The second generation hypoxic cell radiosensitizers SR 2508 (etanidazole) and Ro 03-8799 (pimonidazole) are now undergoing Phase III clinical trials and are predicted from in vitro, animal, and clinical studies to be equivalent to an increase in the tolerable dose of Miso of 5-6 fold. Nonetheless, this will only produce an SER of the hypoxic cells of approximately 1.5 when given with each dose of a 24 fraction course of radiotherapy. There is ample opportunity, therefore, for a sufficiently improved third generation radiosensitizer to warrant the considerable time, effort, and money that would be involved for its clinical testing. Many groups are now involved in the synthesis and testing of drugs with the goal of producing this third generation sensitizer. The purpose of the present review is to summarize what we have learned from the many studies of new radiosensitizers and attempt to draw conclusions about the most fruitful directions for further work. One of the conclusions is that in vitro testing of radiosensitizers, essential though it is as a first screen, can lead to many false predictions. Fortunately, these in vitro systems tend to "overpredict" (i.e., produce false positives rather than false negatives) for in vivo activity. There is only one class of drugs, so far, (those that increase tumor oxygenation) which radiosensitize tumors in vivo, but not hypoxic cells in vitro. Another major conclusion is that the physico-chemical characteristics of drugs that have been shown to be important in enhancing radiosensitization or reducing toxicity (electron affinity, pKa, partition coefficient, DNA binding), have not been systematically optimized in any one compound. Systematic rational drug design is needed to achieve this. The search for new hypoxic cell radiosensitizers must not detract from the fact that a sensitizer of aerobic cells to low radiation doses is needed. A major challenge in achieving a useful drug for radiotherapy is the desirability for such a drug to be tumor specific. We present data which show that tumor hypoxia might be exploited to achieve this goal.