[en] Human exposure limits for laser radiation are based upon experimental ocular injury studies. The limits are derived by committees of ophthalmic experts through a review of all available threshold data and an understanding of mechanisms of laser/tissue interaction. A major point of discussion in this derivation process relates to the level of uncertainty of the threshold of injury. An indication of the level of uncertainty relates to the slope of the transformed dose-response curve, or the 'probit plot' of the data. The most important point on the probit plot is the exposure which represents a 50% probability of injury: the ED-50. It is this value that is frequently referred to as the 'threshold', even though some experimental damage points exist below this 'threshold'. The slope is related to the reciprocal of one standard deviation of the normal distribution of experimental data and thus it reflects not only natural biological variation, but also the impact of experimental errors. The class of damage mechanism will also alter the steepness of the probit plot. When the steepness is less, it may indicate problems in conducting the experiment. The techniques of probit analysis come from toxicology, and certain inherent assumptions are carried over to laser safety studies. An analysis of a number of example data sets reveals that the slope in some experiments could not be explained by biological variation alone. This type of critical analysis is essential in deriving exposure limits. For example, if the slope is not very steep, as with some retinal injury studies, the probit curve may suggest that at one-tenth the ED-50 energy value, there might be a 0.1% risk of injury-a risk generally not acceptable in the laser safety community. Yet, from fundamental biophysical principles, this result could be shown clearly to be flawed. If the ED-50 energy corresponds to a retinal temperature elevation of 15 deg, an energy of 10% of the ED-50 must correspond to 1.5 deg (10% of the ED-50 temperature elevation), which could not produce photocoagulation. This aptly illustrates that any probabilistic risk analysis for laser-induced injury requires one to estimate the true biological variation and separate this from the added experimental errors which reduces the probit slope. Analysis of reported experimental data indicates that the thermal and thermoacoustic damage mechanisms apparently have an intrinsic slope of approximately 1.15 to 1.2. However, experimental threshold data from retinal studies give slopes that are often much greater (e.g., 1.5-1.7), which is really not surprising. The enormous difficulty of seeing a minimally visible lesion and focusing the laser beam to produce the nearly diffraction-limited image leads to this greater spread of data and shallower slopes. If a probit curve is applied to probabilistic risk analyses, it should have a slope of 1.2 or less with the ED-50 point shifted to a lower value. (author)