The usefulness of in vitro systems to predict acute inhalation toxicity was investigated. Nineteen substances were tested in three-dimensional human airway epithelial models, EpiAirway™ and MucilAir™, and in A549 and 3T3 monolayer cell cultures. IC(50) values were compared to rat four-hour LC(50) values classified according to EPA and GHS hazard categories. Best results were achieved with a prediction model distinguishing toxic from non-toxic substances, with satisfactory specificities and sensitivities. Using a self-made four-level prediction model to classify substances into four in vitro hazard categories, in vivo-in vitro concordance was mediocre, but could be improved by excluding substances causing pulmonary edema and emphysema in vivo. None of the test systems was outstanding, and there was no evidence that tissue or monolayer systems using respiratory tract cells provide an added value. However, the test systems only reflected bronchiole epithelia and alveolar cells and investigated cytotoxicity. Effects occurring in other cells by other mechanisms could not be recognised. Further work should optimise test protocols and expand the set of substances tested to define applicability domains. In vivo respiratory toxicity data for in vitro comparisons should distinguish different modes of action, and their relevance for human health effects should be ensured. Copyright © 2012 Elsevier Ltd. All rights reserved.