In the last two decades a number of health status index models have been developed for assessing the value of health outcomes in terms of quality-adjusted life years. The models can be tested by comparing their implications with direct observations of how societies think resources should be distributed across patient groups. This paper reviews empirical evidence of this kind from various countries and summarizes the evidence in three rules of thumb for selecting values for health states. Nine different models are judged relative to these rules of thumb. Eight of the models underestimate the strength of social preferences for treating the severely ill before the less severely ill. The ninth has a strong bias against states associated with emotional distress. As a consequence, none of the models can be seen as sufficient stand-alone instruments for valuing health outcomes. Instead, the models may be seen as complementary and adjustable parts of a tool kit that should also include the rules of thumb suggested in this paper.