OBJECTIVES: The objective of the study is to derive an algorithm to predict a cohort preference-based SF-6D index using the eight mean health dimension scores when patient level data is not available. METHODS: Health related quality of life data (n=6890) collected from patients with a wide range of health conditions was used to explore the relationship between the SF-6D and the eight dimension scores. Ordinary least square regressions were derived using the eight dimension scores and first order interactions. Models were assessed for goodness of fit and predictive abilities using standard statistics such as variance explained; residuals and the proportion of predicted values within the minimal important difference. The models were also compared on their abilities to predict mean cohort SF-6D scores using mean dimension scores using both within-sample and out-of sample published datasets. RESULTS: The OLS equations obtained explained over 83% of the variance in the individual SF-6D scores. While the models over-predict the lower health states and under-predict the higher SF-6D scores on the individual level, the mean absolute errors are in the region of 0.040. When using mean dimension scores from within-sample subgroups and out-of sample published datasets, the majority of predicted scores were well within the minimal important difference (0.041) for the SF-6D. The models are reasonably accurate at predicting incremental values between study arms (mean error 0.012; mean absolute error 0.017) and when predicting incremental changes over time (mean error 0.004; mean absolute error 0.024). CONCLUSION: This paper presents a mechanism to estimate a mean cohort preference-based SF-6D score from published mean dimension scores. This study is unique in that it uses published mean statistics to validate the results. The out-of sample validation demonstrates the algorithms can be used to inform both clinical and economic research. Further research is required in different health conditions.