Under certain circumstances, Navy divers breathe 100% O2 when working underwater. Serious symptoms of central nervous system (CNS) O2 toxicity can develop from hyperbaric O2 exposure; immersion and exercise are also known to exacerbate toxicity. We developed risk models for quantitative prediction of the probability of developing symptoms using a large set of human data in which occupational exposure conditions were simulated. Exposures were 5 to 265 min at PO2 levels from 20 to 50 feet of sea water (fsw) (1 fsw = 3.06 kPa). Approximately half of the exposures were to a single PO2, while the remainder were more complicated consisting of exposures to multiple levels of hyperbaric O2. In 688 trials, there were 42 exposure-stopping symptoms. We used maximum likelihood to estimate parameters, likelihood ratios to compare model fits, and chi 2 tests to judge goodness-of-fit of model predictions to observations. The modeling shows that risk has a steep PO2 dependence. A model with autocatalytic features fits the data as well as a simpler model: when PO2 is elevated beyond 34 fsw, risk accumulates rapidly without bound while accumulating toward an asymptote at lower PO2 levels. This autocatalytic feature of risk accumulation implies a testable hypothesis that substantial protection from human CNS O2 toxicity can be obtained from intermittent exposure (periodic exposure to lower PO2). The models predict that the probability of O2 toxicity is less than 7% with current Navy limits while breathing 95% O2. Probability of symptoms is < 1% if FIO2 is maintained at the United States Navy recommended level of 75%.