Membrane fouling is inevitable for liquid membrane-filtration processes. While the relationship between surface free energy and membrane fouling is now well-established for aqueous feeds, such a link remained missing for the emerging organic solvent applications. In this study, three organic solvents (namely, methanol, ethyl acetate and acetonitrile) and three colloidal foulants (namely, silicon dioxide (SiO 2), titanium dioxide (TiO 2) and aluminum (Al)) were investigated using the same polyacrylonitrile (PAN) ultrafiltration membrane at five transmembrane pressures (100–500 kPa). The zeta potentials were measured, and the Gibbs free energy of interfacial interactions for both foulant – membrane and foulant – foulant were quantified via the DLVO and XDLVO models to correlate with the fouling trends. Surprisingly, although the XDLVO-based interaction energy is generally acknowledged to correlate well with membrane fouling trends for aqueous feeds, the results here indicate that qualitative agreement with flux decline, cake resistance and cake compressibility was poor. Zeta potentials also exhibited poor predictive capability. Only the DLVO-based interaction energy values agreed with the flux decline trends, but nonetheless failed for cake compressibility trends. This study underscore the different fouling behaviors between aqueous versus organic solvent environments, particularly in the limited capacity of the XDLVO model and zeta potential values to provide insights on the fouling mechanisms. • Flux decline trends agreed with DLVO-based foulant-membrane interaction energy. • Some cake resistance agreed with DLVO-based foulant-foulant interaction energy. • Cake compressibility disagreed with foulant-foulant DLVO, XDLVO and zeta potentials. • XDLVO known to predict fouling well for aqueous, but did poorly for organic solvent. • DLVO provided better membrane fouling predictability than XDLVO or zeta potential. [ABSTRACT FROM AUTHOR]