The measurement of bronchial reactivity in epidemiological studies has the advantage of quantifying an objective physiological feature of asthma. Bronchial reactivity was developed in a clinical setting and has been conventionally expressed as the dose of agonist producing a 20% fall in FEV1 (PD20). As PD20 can be estimated for less than 20% of subjects in general community surveys with the doses of agonist that are usually given, data from most subjects must be censored. Thus PD20 alone is a poor index of bronchial reactivity for epidemiological studies. Data from 809 aluminium smelter workers were used to evaluate alternative methods of analysing bronchial reactivity. Dose-response relationships were analysed by four methods: (1) PD20 by the conventional method of interpolating the dose on a logarithmic scale between the last two measurements of FEV1; (2) PD20 (with allowance for extrapolation), estimated by fitting an exponential curve to the dose-response data; (3) the linear regression slope between dose and FEV1 when significant; (4) the dose-response slope obtained in all subjects as the % change in FEV1 from baseline in response to total dose. When each of these measures was related to symptoms, diagnosis, and treatment of asthma, all differentiated between 'asthmatic' and 'non-asthmatic' subjects. The dose-response slope (method 4) had the advantages of simplicity and no censored data, and was shown to be clinically relevant. It is suggested that the dose-response slope should be used for the analysis of bronchial reactivity in epidemiological studies.