The accuracy of exposure data from surrogate sources such as spouses and colleagues was estimated in an historical cohort of 10,059 metal workers. In a 2 year period from 1986 to 1988, 118 subjects who, in 1986, answered a questionnaire on occupational exposures and smoking habits had died. In 1988 spouses and long-term colleagues were interviewed on the same items. Separate contingency tables were developed comparing case and spouse/colleague responses. Percentage of concordance, sensitivity, specificity, the kappa measure of agreement, and a bias factor (exposed to nonexposed ratio between surrogates and cohort members) were calculated. Compared with the index subjects indications, the colleagues' and spouses' reports represented a moderately high level of concordance on occupational exposures, primarily for exposures with a moderate prevalence. There was a considerable under-reporting on both occupational exposures and smoking habits expressed in low values of sensitivity and bias factor values below one. The described misclassification of exposure data by using surrogate information might seriously influence the risk estimation and introduce bias.