The reform of health information systems has been made a priority by health managers, public health specialists and technocrats. While each of these groups has promised major benefits from improvements in information systems, insufficient attention has been paid to the limitations placed upon the theoretical possibilities of information technology by the characteristics of the health system of which the information system is but a part. Managers anticipate improved efficiency and rational allocation of resources, but rational decision making does not automatically follow from improvements in information. Epidemiologists and public health specialists seek more effective and equitable health systems but methodological problems and the expense of many conventional epidemiological approaches continue to limit the usefulness of disease surveillance, programme monitoring and evaluation. Both managers and epidemiologists are confronted with the conflicts which arise in seeking to create locally sensitive information systems within centralised health systems. Technocrats see microcomputers as essential for information systems to be truly effective and as a means of liberating health workers from the drudgery of form filling. However, the rate of organisational evolution in the health system has not kept pace with the rapid development of information technology. There are good prospects for considerable health gain to be wrought from reforms in health information systems but to realise these it is necessary that this process be 'action-led' rather than, as is conventional, 'data-led'. The latter approach sees data as the end in itself; the 'action-led' approach, in contrast, regards information as needs to interventions with a focus on how information will influence decisions. For improvements in information to result in improved health, strategies must be adopted which will ensure that information routinely informs decisions and is seen as a means to the end of improving health.