A study of respiratory morbidity among kaolin (1318747) workers and associated criticism of the survey are reviewed. Objections are raised concerning the heterogenicity of the comparison population, selection criteria, inclusion of former workers in the cohort, and analysis of lung function used. Availability of spirometry data on normal reference populations has been an important part of modern research. This data is not substituted for selected comparison groups. In the kaolin study, the comparison population was selected for demographic, ethnic and occupational characteristics matching the study population. The socioeconomic match was better than would have been given by standard population equations. Study personnel were selected by the company employing the workers. Management was careful to avoid selective recruitment bias. Inclusion of former workers did not bias results; removing them from the analysis caused little change in original findings. Adjustment of outcome variables is a standard procedure in the case of existing confounding factors. Adjusted lung function was consistently lower in all kaolin/comparison group contrasts. The authors conclude that disparity between different studies is not uncommon in epidemiologic research. Technique and methods, as well as the cohorts themselves, can differ. In addition, a large degree of random variability is found in epidemiology. The combination of low statistical power and small samples may have important effects not detected or detected in one study and not another.