OBJECTIVES: The EQ-5D-3L descriptive classification defines a total of 243 health states which presents a problem when seeking to establish social preferences. As it would be challenging to value all 243 health states, a subset is chosen but the basis for this selection varies across national valuation studies. The aim of this study was to choose health states based on the most commonly found health states experienced by the Irish population. METHODS: EQ-5D data from four different datasets were combined to determine what health states are prevalent in Ireland.  Data from a general population study of health (SLAN), an over 70 population cohort, a rheumatoid arthritis and psoriatic arthritis cohort.  The most commonly experienced health states were determined and these were arranged on a 5 dimensional lattice.  Health states were chosen using the Manhattan distance metric.  RESULTS: 12,520 ratings of self-reported EQ-5D health states were included. Fifty two per cent of the cohort had perfect health (11111).  Ninety five per cent of states include at least one '1' and no '3'. 126/243 health states were not experienced in these datasets. The Manhattan distance between health states was measured.  The imposition of such a metric facilitated the identification of clusters of states and associated centroids. Distance sampling was used to identify states within the clusters.  A simple random sampling strategy was also used across the lattice to ensure coverage of health states outside of the cluster CONCLUSIONS: Previous population preference elicitation studies have used theoretical approaches to health states elicitation, which could lead to health states being directly valued which are rarely experienced in the population.  The approach presented here uses the information already known about the population, to inform choice of health states for population valuation of health using the EQ-5D.