OBJECTIVES:  EQ-5D-3L is widely used to generate the utility part of QALY (Quality adjusted life years).When valuing health states of the EQ-5D, a subset of health states with empirically determined values is used to estimate the utility of all possible health state. Clear guidance is lacking on what subset of health states is most suitable for direction valuation. In literature on experimental design issues, different desirable properties have been identified which cannot all be satisfied at the same time and no design is universally the best.Hence, researchers take a leap of faith when estimating a value set in the sense that the suitability of their design is unknown. In this study, we explore the use of orthogonal design as guidence in selecting health states and make comparison with designs used in previous studies. METHODS:  126 students is Erasmus University Rotterdam valued all 243 EQ-5D-3L health states using VAS(Visual Analogue Scale). Next, we examined different orthogonal and non-orthogonal designs in predicting the values of health states. To makre comparison with orthogonal desin, we used the samples of the 42 MVH states, the 17 Dutch/Japanese states and 25 Paris states. Stability was tested by changing the orthogonal coding and by reducing the observations by half. RESULTS:  Among all the designs, an orthogonal design with 54 health states gave the lowest MAEs. Tweaking the orthogonal design to include more common health states had no obvious effect. The main effect model outperformed the N3 model in all designs. The MVH, Dutch/Japanese and Paris samples are suboptimal, compare to orthogonal designs.The result of orthogonal design remains stable when change the orthogonal coding or reduce the observations by half. CONCLUSIONS:  The orthogonal design appears to be a promising health-state selection method for EQ-5D-3L valuation studies. Future studies should test its performance in time trade-off values and for the EQ-5D-5L system.