With the development of cosmetic industry new products arrive to the market on a daily basis. Added fragrances are one of the main reasons why a customer will buy the product, but at the same time, these fragrances may cause irritation of the skin and hypersensitivity reactions. In this research, we focused on 8 fragrances that can be found in cosmetic products and their autoxidation products, which have even greater potential for causing allergic reactions than the pure fragrance which does not contain impurities and autoxidation products. In silico methods are new alternative methods to previous in vivo tests on animals, which determine the toxicological profile of substances based on their physico-chemical properties and structural alerts and presume that similar chemical compounds exhibit similar activity. We used three computer programs to predict irritation and sensitization potential of selected fragrances: Toxtree, Derek Nexus and VegaNIC. We compared the obtained predictions with results obtained from in vivo studies on humans and animals, which helped us to estimate predictive strength of used programs. When predicting skin irritation, the programs Toxtree and Derek Nexus showed contradictory results, however after examining the literature, we determined that Derek Nexus prediction is in fact more accurate. The programs seemed to give more consistent results when predicting sensitization potential. The programs showed unanimous and also accurate predictions for geraniol, hexylcinnamal in butylphenylmethylpropional (BMHCA), while alpha-isomehylionone was predicted sensitization potential, which is not in line with results obtained in studies on animals and humans. In structures of linalool, citronellol and limonene Toxtree did not find structural alerts for skin sensitization while the other two programs classified them as a skin sensitizers. Most in vivo studies identified pure linalool, citronellol and limonene either as weak or non-sensitizers while their oxidation products were skin sensitizers. Furthermore, it is hard to determine which computer program has a more accurate predictive power due to the contradicting results compared to those obtained from the in vivo studies. Nevertheless, Derek Nexus and VegaNIC show a greater predicting power compared to that of Toxtree. In silico methods are a good alternative for avoiding time-consuming, expensive and unethical in vivo tests, however, a lot more has to be done, especially in terms of building databases and improvement of QSAR models.