A new apparatus is described for the measurement of the octanol–water partition coefficient, K_ow; it incorporates features of the slow-stir, filter-probe and generator column methods. Drops of the aqueous phase are passed through a column of octanol solution and the approach to equilibrium is monitored by following the absorbance of the aqueous phase. Experiments were run with 13 acrylate esters; the group includes mono-, di-, tri- and tetra-acrylate esters and spans a large range of hydrophobicity. At equilibrium, the octanol phase was analysed for the ester and water, and the aqueous phase was analysed for the ester and octanol. For 12 compounds, the experiments were run at various concentrations and in all cases the concentration of ester in octanol is a non-linear function of the concentration of ester in water. The non-linearity can be explained by dimerisation of the ester in the octanol phase for 10 of the 12 compounds (group 1); the range of log K_ow is −0.2 to 3.1. For two hydrophobic compounds (group 2), the non-linearity can be explained, in part, by tetramerisation of the ester in the octanol phase; the log K_ow of these is 5.5 and 6.1. At the end of partitioning experiments, the molar absorption coefficients of group 1 correspond well with molar absorption coefficients of aqueous solutions that were prepared separately. In contrast, the molar absorption coefficients of group 2 are much higher than expected and are interpreted as measures of turbidence; this is evidence for the existence of micelles in the aqueous phase. Experiments with stearyl acrylate, the most hydrophobic ester of the thirteen, at a single concentration in the octanol phase show a strong dependence of the partition coefficient on the concentration of octanol in water. This may be interpreted by a micelle mechanism involving both octanol and the ester. Three esters form a series of the general structural form R₁-(CH₂CH₂O)_m-R₂ with m=1, 2 and 3. The partition coefficient decreases with the increasing number of ethyleneoxy moieties, m; the inclusion of one moiety decreases log K_ow by 0.1–0.2, in agreement with other work.