Economical, ethical and scientific constraints have motivated the scientific community to develop alternatives to animal testing. For chronic and systemictoxicity testing, which consume a large number of laboratory animals, no alternatives are available. Developing alternatives in this area representsa challenge because of the complex biological processes implied. A realistic approach could rely on the combination of data generated for multipleendpoints. The Ctox panelstudies conducted in a blinded manner indicated that the method had good sensitivity and specificity (91% and 78%) while defining a LD50 thresholdat 2000 mg/kg. However, the model failed to accurately predict very toxic chemicals displaying a LD50 value below 300 mg/kg. In order to improve® which is a multiparameter, cell-based in vitro system for predicting rat acute systemic toxicity is a typical example. Preliminarythe algorithm, an in depth review and analysis of the false negative and false positive predictions was performed. Conclusions drawn impliedconsiderations of both pharmacological (for the reduction of false negatives) and physical-chemical properties (for the reduction of false positives).The modified approach was applied to 76 non-proprietary compounds previously tested with that statandard method. The predictive performance ofthe model for the Globally Harmonized System (GHS) categories, was assessed. A significant improvement in the prediction of the GHS categories wasobserved. Indeed, 75% of the chemicals pertaining to GHS 1, 2 and 3 were correctly classified, compared to 50% with the standard model. In addition,at an LD50 threshold of 500 mg/kg, the sensitivity and specificity were 85% and 89% with the new model against 71% and 83% with the standard model.Future directions will consist of challenging the newly built model with a new set of chemicals.