A prevalidation study sponsored by the European Centre for the Validation of Alternative Methods (ECVAM) on in vitro tests for acute skin irritation is aimed at identifying non-animal tests capable of discriminating irritants (I) from non-irritants (NI), as defined according to European Union and OECD. This paper reports on Phase III for one of the methods, the skin integrity function test (SIFT), assessing the protocol performance of the SIFT, in terms of reproducibility and predictive ability, in three laboratories. The barrier function properties of excised mouse skin were determined using a set of 20 coded chemicals (10 I, 10 NI), using the endpoints of trans-epidermal water loss (TEWL) and electrical resistance (ER). The basis of the SIFT prediction model is if the ratios of the pre- and post-application values for either TEWL or ER are greater than five-fold, then the test chemical is deemed irritant (I). If the ratio of both parameters is less than five-fold then the chemical is deemed non-irritant (NI). Analysis of variance (ANOVA) indicated that the intra-lab reproducibility was acceptable but that the inter-lab reproducibility was not. Overall, the SIFT test under-predicted the irritancy of the test chemicals chosen for Phase III with an overall accuracy of only 55%. The sensitivity value (ability to correctly predict I) was only 30%. The specificity (ability to predict NI) of the test was better at 80%. A retrospective examination of the SIFT results was undertaken using Student's t-test and a significance level of P<0.05 to predict an irritant based on changes in the TEWL ratio values. This improved the predictivity of the SIFT test, giving a specificity of 60%, a sensitivity of 80% and an overall accuracy of 70%. Appropriate modifications to the prediction model have now been made and the SIFT will be re-examined in a new validation exercise to investigate the potential of this non-animal method to predict acute skin irritation potential.