Infinite-dilution activity coefficients (IDAC) can be used to predict, for example, the behavior of liquid-liquid equilibrium or to determine parameters for excess Gibbs free energy expressions like NRTL, Wilson, or UNIQUAC. For systems where limited (or no) experimental data are available, predictive tools as the well-known UNIFAC variations can be very helpful. An alternative approach is to use methods based on conductor-like screening model (COSMO). In this work, we have tested a modified UNIFAC model with a recent parameter set and three variations of the COSMO-SAC model. To compare the models, a database with a total of 748 IDAC experimental points was assembled with data from the literature and is available as Supporting Information. For nonaqueous solutions, the IDAC logarithm absolute average deviation for UNIFAC was 0.28, while it was 0.48 for a COSMO-SAC with parameters adjusted in this work. On the other hand, for aqueous systems, a COSMO-SAC with optimized parameters led to a deviation of 0.81 against 1.47 for the UNIFAC model tested.