Because airborne exposure varies greatly over time and between individual workers, occupational hygienists should adopt sampling strategies which recognize the inherent statistical nature of assessing exposure. This analysis indicates that the traditional practice of testing ‘compliance’ with occupational exposure limits (OELs) should be discarded. Rather, it is argued that acceptable exposure should be defined with reference to the exposure distribution. Regarding the many statistical issues which come into play, it is concluded that hygienists should continue to apply the log-normal model for summarizing and testing data. However, sampling designs should move away from methods which are biased (e.g. sampling only the worst case) and which rely upon job title and observation as the primary means of assigning workers into groups. Since exposure data often lack independence (e.g. owing to the autocorrelation of serial measurements) and there exist large differences in exposure between workers in the same job group, random sampling designs should be adopted. It is also shown that the relationship between the mean of a log-normal distribution and exposures in the right tail allows one to evaluate simultaneously the mean exposure and the maximum frequency with which exposures exceed the OEL.

Investigation of the biological concepts relies heavily upon a conceptual model which depicts the exposure-response continuum as a sequence of time series related to exposure, burden, damage and risk. Analysis of the linkages between these processes identifies two kinetic conditions which are necessary if variability of exposure is to affect appreciably the individual's risk of chronic disease. First, the variation of exposure from interval to interval must be efficiently translated into burden and damage (no damping), and second, during periods of intense exposure the relationship between burden and damage must be non-linear (curving upwards). On the basis of current knowledge it appears that relatively few chronic toxicants satisfy both these conditions. Even for those substances which cause damage only when a threshold is exceeded, a statistical argument suggests that the maximum risk can still be related to the mean exposure received over time.

It is concluded that the risk of chronic disease generally depends upon the mean exposure received by the individual worker over time. Thus, the sampling strategy must allow the distribution of individual mean exposures to be characterized across the population at risk. It follows from this paradigm for assessing exposures that relatively little effort should be devoted to the evaluation of short-term ‘peak’ exposures since such transients are unlikely to exert undue influence on long-term effects. Regarding short-term effects it is argued that evaluation of exposures to acutely toxic agents should rely upon assessment and control of the source rather than of exposure per se.

Finally, three approaches are evaluated for testing exposures relative to limits: the test of compliance; a test of the frequency of excursions; and a test of the mean exposure: By several objective N criteria it is shown that the traditional test of compliance offers a poor tool for prospective assessment. While both of the other approaches allow rigorous testing to be conducted, evaluation of the mean1 exposure appears superior on the grounds of statistical power and the relationship between outcome and the long-term hazard. However, investigation of the underlying bases for OELs indicates that exposure limits can allow more risk to prevail than is generally presumed and the variation of exposure among workers in a group can be large. Thus, it may be necessary to incorporate safety factors into OELs prior to testing mean exposures.