In 1996 and 1997, ECVAM supported a formal validation study on in vitro methods for predicting skin corrosivity. Two of the in vitro tests included in the study employed human skin models, the Skin2™ ZK1350 and EPISKIN™ models. In the ECVAM validation study, BASF, Huntingdon Life Sciences (HLS) and ZEBET tested the Skin2 human skin model, production of which ceased in October 1996, while the validation study was still in progress. Since both of the skin models had shown basic usefulness for corrosivity testing and, in particular, the EPISKIN corrosivity test had proved to be a scientifically valid test, the three laboratories decided to conduct a study to determine whether another commercially available human skin model, EpiDerm™, could also be successfully used to predict skin corrosivity. The study was performed according to the ECVAM prevalidation scheme, to allow for refinement of the test protocol and the prediction model, as well as for independent assessment of the performance of the refined methodology in a final blind trial in the three laboratories. In phase I of the study, ZEBET (Laboratory 1) drafted a Standard Operating Procedure (SOP), including a prediction model (PM1), and the project plan for the study. It was a major task to simplify an existing EpiDerm test protocol, which used the time-course of cytotoxicity as its endpoint. To evaluate the predictivity of the simplified method, which used only a 3-minute exposure to test chemicals, 50 chemicals representing a wide spectrum of chemical entities were tested, revealing that the test sensitivity was too low (65%), whereas the specificity was very high (88%). In addition, acceptance criteria for the negative and positive controls were established. Before proceeding to the next phase of the study, ZEBET distributed a refined SOP, data-recording software and documentation sheets, which allowed Good Laboratory Practice (GLP)-compliant quality assurance for each assay. The main goal of phase II was to produce sufficient data to assess the reproducibility of the EpiDerm skin corrosivity test after transfer to Laboratory 2 (HLS). Repeated testing of several chemicals in both laboratories revealed excellent intralaboratory and interlaboratory reproducibility. In addition, chemicals classified as "non-corrosive" (NC) with a 3-minute exposure in phase I, were re-tested by ZEBET with extended exposure periods of 1 hour and 4 hours. The test sensitivity could be significantly increased, if chemicals classified NC with a 3-minute exposure were tested with a 1-hour exposure. Before proceeding to the final blind trial, a refined SOP was drafted, according to which all chemicals had to be tested with exposure times of 3 minutes and 1 hour, and data for these two exposure times were used in the refined hierarchical prediction model, PM2. In phase III, the blind trial, BASF (Laboratory 3) joined the study. ECVAM selected 24 chemicals from the test chemical set used in the ECVAM skin corrosivity validation study, and BIBRA International (UK) purchased, coded and distributed the chemicals. Each chemical was tested twice, independently, according to the principles of GLP, and coded data were submitted to the Humboldt University (Berlin, Germany) for biostatistical analysis. The analysis revealed that the final test protocol and the refined prediction model (PM2) provided a highly balanced prediction of 88% sensitivity and 86% specificity, which is regarded as the best predictivity an in vitro skin corrosivity test can be expected to achieve. In conclusion, the EpiDerm skin corrosivity test gives an excellent prediction for a wide spectrum of chemicals, and could be used within the context of the new Annex V (EU Dangerous Substances Directive) test method (human skin model assay) for skin corrosion. The results obtained were reproducible, both within and between laboratories, and showed that EpiDerm could be used for testing a wide range of chemicals (both liquids and solids), including organic acids and bases, neutral organics, inorganic acids and bases, electrophiles and phenols. The concordances between the skin corrosivity classifications derived from the in vitro data were very good, and the test was able to distinguish.