Few workplace health promotion (WHP) interventions are designed to improve work conditions. Methods for measurement of work conditions are often developed from a risk factor perspective rather than a WHP perspective. More knowledge is needed on the work conditions that promote health in order to develop a good work environment. The purpose of the present study was to investigate if the Demand Control Support model, the Effort Reward Imbalance model and the Job Characteristic Inventory are correlated, if the subscales predict health and to analyze which combination of subscales is the most useful predictor of health longitudinally. The study used questionnaire data from 662 civil servants at baseline and at follow-up 2 years later. The data were analysed by multiple regressions. A new model; effort, reward, and variety, was found having a higher predictive power to predict health than the original models. To promote health at work, social relations and health-mediating work conditions are important because these conditions may buffer health. Health can be assumed to be a resource that is created in everyday activities and interactions in workplaces, and there is a need to develop health measure instruments based on holistic health theories.