The separate and combined effects of duration and intensity of exposure to crocidolite on mortality from lung cancer, malignant mesothelioma, and stomach cancer were examined in 6506 male former crocidolite miners and millers at Wittenoom Gorge, Western Australia. Each subject who had died from lung cancer (92), mesothelioma (31), or stomach cancer (17) was matched with up to 20 control subjects of the same age who were not known to have died before the index subject. Relations of dose and time of exposure to crocidolite to risk of death were modelled by conditional logistic regression. For lung cancer, the best fitting multiplicative model was one which estimated a relative risk (RR) of 1.12 (95% CI 1.04-1.20) per year of exposure and 1.01 (95% CI 1.00-1.01) per fibre/ml. This was statistically indistinguishable from an additive model showing an increase in RR of 0.01045 (95% CI 0.008-0.020) per f/ml year. For mesothelioma the best fitting model appeared to be one estimating a RR of 24.9 (95% CI 3.51-1.77) per log year since first exposed and a RR of 10.5 (95% CI 3.12-35.1) if exposed for longer than six months. This was not distinguishable statistically from a model that showed mortality increasing as the fourth power of time since first exposed less the fourth power of time since last exposed. The effect of intensity of exposure on the RR for mesothelioma was only slight. There was no consistent effect of any measure of exposure to crocidolite on death from stomach cancer.