CAM-based assays, in which test material is applied to the chorion allantoic membrane (CAM) of embryonated chicken eggs, were assessed as alternatives to the Draize eye irritation test. Two general types of CAM-based assays are currently in use, the HET-CAM test and the CAMVA assay. Evaluations were made of five data sets produced with three different modifications of the HET-CAM test and two data sets obtained with the same CAMVA protocol. Data sets consisted of 9-133 test chemicals, usually from the sponsor's product line, and also from a validation trial. Each data set and assay protocol were analysed for quality of data, purpose and proposed use of the assay, range of responses covered, range of test materials amenable, current use in safety and risk assessment both in-house and for regulatory purposes. Since the MMAS Draize score was not available for all in vivo data sets, the sigma MMMIS, which correlates well with the MMAS, was used instead. In vitro/in vivo correlations calculated with Pearson's linear coefficient ranged from r = 0.6 to r = 0.9 for six of seven data sets. Corneal opacity and inflammation of the iris showed the best correlation to in vitro data. Prediction rates were significantly improved when partial linear regression was used, and the predictivity of three different HET-CAM protocols was almost the same. HET-CAM assays showed the best prediction with surfactants and surfactant-based formulations, whereas the CAMVA assay provided the best performance with alcohols.