Background In April 1974@ Tabershaw/Cooper Associates@ Inc. was requested by the American Petroleum Institute to conduct a study of illness in petroleum refinery workers. A previous study (1) had evaluated the mortality experience of this occupational group@ and a need existed to extend the assessment of the health of these workers to their illness experience. The study was motivated by the need to describe the risk of illness in the industry@ both to obtain a baseline for future comparisons@ and to examine morbidity in those body system which had been shown to be of particular interest in the earlier mortality study. However@ the measurement of illness in an employed population presents problems both in the identification of illness episodes and in the determination of cause of illness. The problems lie in the development of a data collection system and the development of criteria for the diagnosis and validation of illness. Further@ such developments are not in general applicable on a retrospective basis@ but require a period of data collection under a planned system@ followed by an analysis. In the meantime@ the present management of sickness absenteeism at the plant level does in fact generate records. The possibility exists that such records can be used to generate morbidity statistics@ using definitions and criteria which make use of the existing data@ and are consistent with recommended practice. Even if this is not the case@ the possibility must be explored and evaluated as a basis for any further development of planned morbidity studies. Therefore@ the present study was designed to evaluate the feasibility of obtaining meaningful morbidity statistics from routinely collected data in plant records. 1. To attempt to obtain appropriate data for the calculation of morbidity statistics from a representative selection of petroleum refineries. 2. To document the problems in collecting such data@ from the point of view of their validity and of the difficulties of retrieval. 3. To present and interpret the resulting statistical measures