AIM: The discounted utility (DU) model has dominated economic evaluation for almost 7 decades, despite the fact that important assumptions of the model are commonly found to be violated. This paper formally explores whether the key assumption of stationarity is violated in a sample of the general population of Northern Tanzania. Furthermore, three hyperbolic discounting models are fitted to the data, and whether they perform better than the DU model in predicting individuals' time preferences is tested using nonlinear least squares regression. METHOD: The data were collected from 450 households by trained enumerators. The individual data on time preferences were collected by structured interviews using an open-ended stated preference methodology. Respondents marked a rating scale to indicate the maximum number of days they would be willing to suffer a nonfatal disease if the outbreak of the disease could be delayed to a point further into the future. Households were randomised to answer questions framed to elicit either a private or social time preference. RESULTS: Hypothesis testing confirmed decreasing time aversion and a magnitude effect, suggesting that the DU model is inappropriate as a descriptive tool. When the DU model was compared with the three hyperbolic discounting models by analysing the discount factor using nonlinear least squares regression, the most important findings were that a variable for starting point was nonsignificant only for the Loewenstein and Prelec (L&P) and the Mazur models, and that people in this setting generally discounted future health far more than suggested by current discounting practice in economic evaluations. CONCLUSION: The time preferences of our sample are better represented by the L&P and the Mazur models (which allow relaxation of the stationarity assumption through a modification of the expression for the discount factor) and less well reflected by the Harvey (a modification of the L&P model that assigns more importance to the future than standard utility discounting) and DU models. This implies that, from the point of view of a consumer sovereignty-friendly economist, the Mazur and the L&P models are preferable for discounting of future health in economic evaluations. However, from the point of view of other value bases for discounting the choice of discounting model is of less importance.