In 1993 a national population survey was conducted in the UK to determine social preference weights for EQ-5D health states. The survey, conducted as part of the Measurement and Valuation of Health (MVH) Project was based on face-to-face interviews with 3395 respondents who used ranking, VAS rating and time trade-off (TTO) methods to generate social preference weights. The TTO weights derived from that survey are widely used in economic studies in which EQ-5D is used to measure outcomes. OBJECTIVES: To test the stability of social preference weights over time using a partial replication protocol. METHODS: Face-to-face interviews were conducted with 253 individuals randomly selected from electoral registers in geographically convenient locations. The protocol varied somewhat from the original MVH study in that each respondent was asked to value a standard set of 17 health states. Responses were weighted to achieve matching with the age/gender mix of the MVH study. Fieldwork was conducted in 1998. RESULTS: The rank order of health states was highly similar in the two surveys. VAS ratings were well approximated in the replication survey. TTO weights were higher in the second survey, with fewer states being valued as worse than dead. The median absolute difference in TTO preference weights was 0.080. An OLS model used to interpolate values for unobserved EQ-5D health states yielded lower parameter estimates for the extreme levels on each dimension in the 1998 data. CONCLUSIONS: Social preference weights can and do vary over time. An apparent shift in the value of dead relative to EQ-5D health states may account for a reduction in the number of states assigned a negative weight. Although the absolute value of outcomes may depend upon which set of weights is used, marginal differences in outcome may remain unaffected.