OBJECTIVES: To contrast algorithms commonly employed for predicting health state utility values of multi-morbid health states from single condition health state utility values with multimorbid health state utility values directly measured in the adult population.    METHODS: Cross-sectional regression data analysis of EQ-5D utility values as a function of single, two-condition and three-condition combinations of 12 chronic conditions in the English GPPS data (N~820,000). A linear regression model was used to test for additive, syperadditive or subadditive effects of diads and triads of chronic conditions and to estimate predicted multimorbid state utility values for comparison with the respective utility values predicted by exisiting algorithms based on single condition utility values.      RESULTS: Out of all the 66 possible different pairs of conditions in GPPS, 19 displayed interaction effects that are inconsistent with the multimorbidity state valuation models in the health economic literature. The prevalence of these two-condition combinations is 4.25% in the adult population registered with a GP practice in England, and 18% of those with two or more conditions in this population. Additive models fitted the disutility experience of patients with 35 of the 66 two-condition combinations, accounting for 3.22% of individuals registered with a GP practice in England. In contrast, the minimum model was only found to apply to three pairs of conditions, neurological & asthma or chest, high blood pressure & liver or kidney disease, and cancer and neurological, amounting to 0.17% of the population (and it did not apply to any of 220 possible combinations of three conditions). The rest were chronic condition combinations with utility values that fell between the predictions of the minimum and multiplicative algorithms, and have a prevalence of 5.86% in the adult population. CONCLUSIONS: The algorithms commonly used by researchers to calculate the expected utility value of health states with jointly occurring conditions do not match the observed health utility values of some of the most prevalent self-reported long term condition combinations. The appropriate algorithm for calculating health state utilities of multimorbidity from single condition utilities may differ between condition combinations.