Intertemporal preferences for health are an important concept when modelling health-affecting behaviour and with respect to informing discounting practice in economic evaluation. The aim of this paper is to test robustly stationarity, the key axiom of the Discounted Utility model, and to test whether the quasi-hyperbolic or generalised hyperbolic model provides a better description of individual time preferences for health outcomes when stationarity is violated. Very little is known about the descriptive validity of the quasi-hyperbolic model. The different models can lead to different predictions and it is therefore crucial to test which functional form is more descriptively valid. An axiomatic approach is used. Intertemporal preferences were elicited from 203 university students. The results showed that stationarity is violated. Individuals discounted both initial delay and further delays between outcomes at a decreasing rate. This suggests that the quasi-hyperbolic model may not be appropriate to use in intertemporal analyses of health behaviour.