OBJECTIVES: The use of ordinal preferences based on the rank order of health states has recently attracted attention. A ranking task has traditionally had a limited role as a warm-up exercise prior to more complex preference elicitation methods such as Time Trade-Off. This paper reports on the use of rank data as the primary source of preferences in a health valuation study.  METHODS: 205 university students ranked and rated (using a VAS scale) a subset of 14 EQ-5D health states selected from the US Valuation Survey. The order of presentation was randomised. The logical consistency and number of tied health states were computed for each method. The agreement amongst participants was measured by Kendall’s Coefficient of Concordance (W). Consistency on a within-participant basis was measured by counting the number of states having the same rank as determined for each method. Scale values estimated from the rank data using a conditional logistic regression model were compared with results obtained from a random effects model applied to the VAS ratings.   RESULTS: Seventy-six percent of participants made no logically inconsistent responses in ranking health states. This figure is higher than corresponding 62% recorded for the VAS ratings. An average of 2 states were tied in the ranking task; this figure rose threefold for the VAS rating. Kendall’s W for the ranking task was higher than that for VAS rating. An average of 4/14 states per participant were ranked identically in the two preference elicitation methods. Coefficients based on the rank data yielded results that were similar to those obtained from the VAS data. CONCLUSIONS: Ranking produces data that are logically superior to those obtained from VAS rating. Standard modelling methods enable the recovery of an interval scale for EQ-5D health states. The use of ordinal rank data in deriving health value should be encouraged.