Although double salt ionic liquids (DSILs) offer significant advantages over classical twoion ionic liquids as separation solvents, relevant studies are still scarce and a systematic DSIL selection method is thus highly desirable. In this contribution, a rational method for designing DSILs as extraction solvents is proposed and exemplified by the thiophene/ n-octane separation. The effects of additional degrees of freedom for DSIL design (i.e., double cations and/or anions and the ion ratio) on the thermodynamic properties are first analyzed by COSMO-RS. Then, a multilevel DSIL design method combining the prediction of infinite dilution thermodynamic properties, the estimation of physical properties, the evaluation of phase equilibrium behavior, and the experimental validation is proposed. By applying this method, [C 2 MIm][OAc] x [NO 3 ] 1-x (x = [0, 1]) and [C 2 MIm] [OAc] x [SCN] 1-x (x = [0.70, 1]) are identified as promising DSIL solvents for the thiophene/ n-octane separation. Correspondingly, the liquid-liquid equilibria of {DSILs + thiophene + n-octane} with the designed DSILs are experimentally studied. K E Y W O R D S COSMO-RS, double salt ionic liquids, liquid-liquid equilibrium, solvent design, thiophene/ n-octane separation 1 | INTRODUCTION During the past decades, ionic liquids (ILs) have sparked significant interest for application in various separation processes due to their overwhelming advantages over conventional organic solvents, such as negligible vapor pressure, broad liquid range, designable and tunable character, and so on. 1-10 However, as known to all, various cationanion combinations can lead to very different physical and thermodynamic properties, making the selection of ILs the most crucial task for the development of IL-based separation processes. 4-10 A large number of experimental and theoretical studies have been reported on the identification of suitable ILs for different separation processes. 7-25 Particularly, several research groups have employed COSMO-based activity coefficient models 12-19 or the UNIFAC-IL model 20-25 to perform computational screening and design of ILs as separation solvents. Because of the predictive character of these thermodynamic models, the separation performance of ILs that have not been experimentally covered can be estimated and thus the identification of suitable solvent candidates in a large range is possible. Despite the progress made, the selection of practically attractive IL for a specific process is still challenging as in many cases very few ILs can satisfy different desirable properties simultaneously. For instance, when screening ILs as extractive desulfurization solvent, Song et al found that only 831 of the initial 36,260 cation-anion combinations meet the thermodynamic property requirements, among which only 15 satisfy the physical property constraints. 16 Conceivably, if more restrictions such as easy preparation, commercial availability, low cost, and so on of ILs are considered, the number of feasible solvents will be further reduced...