Respiratory tract toxicity represents a significant cause of attrition of inhaled drug candidates targeting respiratory diseases. One of the key issues to allow early detection of respiratory toxicities is the lack of reliable and predictive in vitro systems. Here, the relevance and value of a physiologically relevant 3D human airway in vitro model (MucilAir) were explored by repeated administration of a set of compounds with (n = 8) or without (n = 7) respiratory toxicity following inhalation in vivo. Predictability for respiratory toxicity was evaluated by readout of cytotoxicity, barrier integrity, viability, morphology, ciliary beating frequency, mucociliary clearance and cytokine release. Interestingly, the data show that in vivo toxicity can be predicted in vitro by studying cell barrier integrity by transepithelial electrical resistance (TEER), and cell viability determined by the Resazurin method. Both read-outs had 88% sensitivity and 100% specificity, respectively, while the former was more accurate with receiver operating characteristic (ROC) AUC of 0.98 (p = .0018) compared with ROC AUC of 0.90 (p = .0092). The loss of cell barrier integrity could mainly, but not fully, be attributed to a loss of cell coverage in 6 out of 7 compounds with reduced TEER. Notably, these effects occurred only at 400 µM, at concentration levels significantly above primary target cell potency, suggesting that greater attention to high local lung concentrations should be taken into account in safety assessment of inhaled drugs. Thus, prediction of respiratory toxicity in 3D human airway in vitro models may result in improved animal welfare and reduced attrition in inhaled drug discovery projects.