OBJECTIVES: To compare the feasibility and reliability of the binary discrete choice experiment (DCE) and  the multiprofile case of best-worst scaling (BWS) techniques in eliciting preferences for the EQ-5D-5L. METHODS: Forty-eight EQ-5D-5L health states (HS) were selected using a Bayesian efficient design and grouped into 24 pairs for DCE tasks and eight sets for BWS tasks (each set has three HS). Participants completed 12 pairs and eight sets in random order.  Time to complete each task was recorded. Participants were asked to rank each HS using a visual analogue scale (VAS). Each participant completed, for a second time, 3 DCE pairs and 2 sets of the BWS tasks randomly selected from the original pairs and sets, respectively. Participants answered questions about the difficulty in imagining EQ-5D-5L HS and in completing the tasks. A conditional probit model was used to estimate latent utilities for the EQ-5D-5L HS which were subsequently rescaled to facilitate the comparison. RESULTS: One hundred persons participated (mean age: 45 years, 66% female, 75% well-educated). Mean time to complete 12 DCE tasks was 7.7 minutes and 10.1 minutes for eight BWS tasks. Some level of difficulty imagining the EQ-5D-5L HS was reported by 70% of participants. Only 13% of the participants reported no difficulty when choosing between two HS or from three HS. The intraclass correlation coefficient (ICC) was 0.53 for DCE tasks and 0.45 for BWS tasks. The agreement with the VAS rankings was 93% for the DCE tasks and 91% for the BWS tasks Although the rescaled latent utility estimates were similar, the 95% CIs were much wider for the estimates based on BWS than for those based on DCE. CONCLUSIONS: The DCE performed much better than the BWS in eliciting preference to EQ-5D-5L HS. BWS might not be an appropriate technique for valuing the EQ-5D-5L