Natural toxins are widely occurring, highly diverse organic compounds produced by, for example, plants or fungi. In predictive environmental fate and risk assessment of organic chemicals for regulatory purposes, the octanol-water partition coefficient (K-ow) remains one of the key parameters. However, experimental data for natural toxins are largely missing, and the current estimation models for K-ow. show limited applicability for multifunctional, ionizable compounds. Thus, log K-ow data were first experimentally derived for a diverse set of 45 largely ionizable natural toxins and then compared with the predicted values from three different models (KOWWIN, ACD/Percepta, and Chemicalize). Both approaches were critically evaluated with regard to their applicability for multifunctional, ionizable compounds. The miniaturized shake-flask approach allowed reliable quantification of pH-dependent partitioning behavior for neutral, acidic, and basic ionizable natural toxins. All the analyzed toxins are rather polar with an average log K-ow < 1 and an observed maximum log K-ow of 2.7. Furthermore, the comparison of experimental data for the neutral form of ionizable toxins with those of commonly used prediction models showed that the latter match the former with only slightly increased errors. The Chemicalize tool gave the best overall predictions for the dataset generated here, with a mean absolute error of 0.49.