We determined the toxicity of mixtures of ethyl acetate (EA), isopropyl alcohol (IPA), methyl ethyl ketone (MEK), toluene (TOL) and xylene (XYL) with half-maximal effective concentration (EC50) values obtained using human hepatocytes cells. According to these data, quantitative property-activity relationships (QPAR) models were successfully proposed to predict the toxicity of mixtures by multiple linear regressions (MLR). The leave-one-out cross validation method was used to find the best subsets of descriptors in the learning methods. Significant differences in physico-chemical properties such as boiling point (BP), specific gravity (SG), Reid vapor pressure (rVP) and flash point (FP) were observed between the single substances and the mixtures. The EC50 of the mixture of EA and IPA was significantly lower than that of contained TOL and XYL. The mixture toxicity was related to the mixing ratio of MEK, TOL and XYL (MLR equation EC50 = 3.3081 − 2.5018 × TOL − 3.2595 × XYL − 12.6596 × MEK × XYL), as well as to BP, SG, VP and FP (MLR equation EC50 = 1.3424 + 6.2250 × FP − 7.1198 × SG × FP − 0.03013 × rVP × FP). These results suggest that QPAR-based models could accurately predict the toxicity of polar and nonpolar mixtures used in rotogravure printing industries.