Abstract Thirty-three test substances comprising household and personal products and a pure surfactant were used in an independent evaluation of the ability of five selected in vitro assays to predict eye irritation potential. The data were compared with historical archived data from rabbit eye irritation tests conducted on the same subsamples. The data were assessed in three ways. The linear correlation for all 33 test substances with in vivo data was poor for each assay. However, the correlation improved greatly for some assays when a group of similar test substances was considered. Analysis of the data by Cooper's criteria suggested that fluorescein diacetate uptake by Chinese hamster V79 cells (V79) and the Microtox test kit were the best of the assays evaluated in discriminating correctly between the irritants and non-irritants that were identified by rabbit eye irritation testing. Some of the data were selected for further analysis because it was possible to predict differences, based on experience learned from animal data on similar substances, between the test substances without reference to experimental data. Neat or more concentrated test substances were predicted to be more irritant than dilutions and an experimental laundry powder formulation containing 10% sodium metasilicate (MTS) was predicted to be more irritant than two analogous non-MTS containing formulations. The experimental data were then compared with the prediction. The in vivo rabbit eye irritation tests and Microtox distinguished correctly between all the test substances in this analysis. Eytex was unable to distinguish correctly between any of the test substances in this analysis. The other assays were able to distinguish between some of the test substances. No single assay performed well in every type of analysis and it is concluded that a battery of assays is required to obtain reliable predictive data. Overall, Microtox, V79 fluorescein diacetate accumulation and 3T3 neutral red uptake were the most promising assays. Further evaluation is necessary and extreme caution should be used in comparing in vitro results with rabbit eye irritation data to make definitive conclusions.