Allergic contact dermatitis (ACD) is a significant safety concern for developers of cosmetic, personal care, chemical, pharmaceutical, and medical device products. The guinea pig maximization test (GMPT) and the murine local lymph node assay (LLNA) are accepted methods for determining chemical sensitization. Recent legislative initiatives in Europe require the development of new in vitro alternatives to animal tests for chemical sensitization. The aim of this project was to develop an in vitro screening method that uses a human skin cell line (HaCaT), chemical reactivity, and gene expression profiling to identify positive and negative responses, to place chemicals into potency categories of extreme/strong (ES), moderate (M), weak (W), and nonsensitizers (N), and to provide an estimate of corresponding LLNA values. The method and processing algorithm were developed from a training set of 39 chemicals possessing a wide range of sensitization potencies. Three cationic metals, chromium (Cr), nickel (Ni), and silver (Ag), were also evaluated in this model. Chemical reactivity was determined by measuring glutathione (GSH) depletion in a cell free matrix. Three signaling pathways (Keap1/Nrf 2/ARE/EpRE, ARNT/AhR/XRE, and Nrf1/MTF/MRE) that are known to be activated by sensitizing agents were monitored by measuring the relative abundance of 11 genes whose expression is controlled by one of these 3 pathways. Final exposure concentrations were based on toxicity and solubility. A range-finding experiment was conducted with each compound to determine cytotoxicity and solubility. Six exposure concentrations (0.1 to 2,500 microM) and an exposure time of 24 hours were used in the final experiments. Glutathione depletion alone did not provide the accuracy necessary to differentiate potency categories. However, chemical reactivity combined with gene expression profiles significantly improved the in vitro predictions. A predicted toxicity index (PTI) was determined for each test chemical. A comparison of LLNA values with PTI values revealed an inverse relationship. The large variation in LLNA data for compounds in the same potency category makes direct extrapolation from PTI to LLNA difficult. To challenge the system, 58 additional compounds were submitted in a blinded manner. Compounds placed into ES and M categories were considered positive, whereas compounds classified as W or N were considered negative. Accuracy was approximately 84%, with a sensitivity of 81% and a specificity of 92%. The model correctly identified 2 of 3 cationic metals as positive. In conclusion, the method described here demonstrates a valuable in vitro method for identifying chemicals and metals that induce skin sensitization.