The multivariate statistical method Principal Component Analysis (PCA) has been applied to a set of data from the ECETOC reference chemical data bank. PCA is a multivariate method that can be used to explore a complex data set. The results of the analysis show that most of the variability in the values for tissue damage scores for the 55 chemicals can be described by a single principal component which explains nearly 80% of the variability. This component is derived by giving approximately equal weight to each of the 18 individual measures made on the tissues over the 24-, 48- and 72-hr observation period. The principal component scores on the first component (PC I) are very highly correlated with the maximum individual weighted Draize scores or total Draize scores (TDS) derived using the Draize scoring method. A second principal component, describing about 7% of the variability, contrasts damage measured on the iris and cornea with that measured on the conjunctiva. Plots of principal component scores show the overall pattern of responses. In general, low measures of the TDS and a positive (PC I) score are associated with iris and conjunctival damage (damage to the iris was never recorded in the absence of damage to the conjunctiva). High TDS and negative PC I scores are associated with corneal and/or iris and conjunctiva damage. Plots of the principal component scores identify some chemicals that appear to cause unusual patterns of damage and identify some individual animals as having outlying or idiosyncratic responses. However, the analysis suggests that (i) there is only limited evidence for differential responses of different tissues and (ii) that attempts to identify alternative tests which predict specific types of tissue damage based on the results collected in a Draize test are likely to be unsuccessful. It indicates that further refinement of the results of the in vivo Draize test will not arise from more detailed analysis of the tissue scores but by refinement in the understanding of the mechanisms associated with the test. PCA was shown to be a powerful statistical tool for the investigation of complex data sets and provides a succinct description of such data sets, allowing patterns to be identified and the potential to develop further hypotheses for investigation.