A novel method based on unimolecular quantum mechanical calculations has been used to predict the multicomponent liquid-liquid equilibria (LLE). The recently developed COnductor-like Screening MOdel (COSMO), along with the most common quantum chemical package of Gaussian03 has been used in this work. COSMO-RS combines the COSMO model calculations with exact statistical thermodynamics of pairwise interacting surface segments and has been used for the evaluation of molecular interactions in liquids. COSMO-RS has been used to predict the LLE for 158 multicomponent data sets, which consist of 80 ternary, 9 quaternary, and 4 quinary systems, several of which are examined at different temperatures. The effective contact surface area (a(eff)), the hydrogen-bonding coefficient (c(hb)), and the cutoff surface charge density for hydrogen bonding (sigma(hb)) have been estimated simultaneously, using 10 ternary data sets that consist of 11 different functional groups (CH3, CH2, CH, CH=CH2, C-CH3, OH, CH2O, C-(CH2), COO, C-Cl, CH2-CN) and 9 different solvents (sulfolane, dimethyl sulfoxide, N-methyl pyrrolidone, furfural, propylene carbonate, triethylene glycol, tetraethylene glycol, ethylene carbonate, dicyanobutane). Cavity radii (r(i)) values up to 1.17 times greater than the Bondi radius have been used. The prediction has been initially benchmarked on the octane-toluene-sulfolane system at three temperatures: 25, 35, 45 degrees C, for which the root-mean-square deviations (RMSDs) are < 1% for mole fraction predictions. Thereafter, the LLE has been predicted for 158 multicomponent data sets. An important aspect of our work is that, for the first time, quaternary and quinary systems have been predicted using COSMO-RS. The overall RMSD is < 5% for multicomponent systems, compared to similar to 1% for NRTL and UNIQUAC models and 5% for UNIFAC models.