OBJECTIVES: The EQ-5D is often used to measure health utilities. Uncertainty in the EQ-5D scoring algorithm is routinely ignored. We aim to quantify the extent of uncertainty in the US EQ-5D-3L scoring algorithm, which was based on data from 3773 respondents -- largest valuation study to date. METHODS: We re-fitted the US scoring algorithm using the same data and functional form as was originally used, omitting each health state in turn and examining the error in the predicted mean utilities. We then used a mixed effects model, including a random effect for health state, adopting a Bayesian perspective to estimate the predictive distribution of the mean utilities for health states not included in the valuation study (which captured 43 of 243 health states). This allowed us to estimate uncertainty in the scoring algorithm. RESULTS: The mean absolute error for predicted mean utilities on cross-validation was 0.033; the mean absolute error for a perfect model, accounting for sampling error in the observed mean utilities, would have been 0.01. The root mean squared error was 0.042; for a perfect model it would have been 0.013. The standard deviation for the random effect for health state was 0.03, suggesting that the width of the confidence interval for the mean utility for a randomly selected health state is around 0.12. The Bayesian model indicated that the width of the 95% credible interval for the mean utilities varied from 0.015 to 0.45, with a median width of 0.18 and interquartile range of 0.15 to 0.22. CONCLUSIONS: EQ-5D health utilities are subject to considerable uncertainty (for comparison, the MID for EQ-5D utilities is 0.05 to 0.08). Other countries’ scoring algorithms are based on smaller sample sizes and so subject to greater uncertainty. This uncertainty should be accounted for when using EQ-5D health utilities in economic evaluations.