In the development of new drugs, it is very important to know the effects these may bring to those who consume them. Drugs which act upon certain diseases must not cause toxic side effects on healthy organs. These toxic side effects can be quite varied, i.e. mutagenicity, clastogenicity, teratogenicity, etc., but undoubtedly the mutagenicity officiate in the selection process, during preclinical testing, to advance in clinical trials. Mutagenic compounds are removed and cannot continue its development. There are preclinical studies of mutagenicity and genotoxicity, ranging from in vitro to in vivo studies. Particularly, Ames test is recommended by ICH as the first input in these studies. Herein, we investigated the mutagenicity of an in-house chemical library of eighty five N-oxide containing heterocycles using Ames test in Salmonella thyphimurium TA 98 with and without S9 activation and the use of neural networks in order to predict this nondesired activity. N-oxide containing heterocycles are especially relevant regarding its pharmacological activities as antitrypanosoma, anti-leishmania, anti-tuberculosis, anti-cancer, chemopreventive, anti-inflammatory, anti-atherogenic, and analgesic agents. In some cases, a relationship was found between the presence of N-oxide and mutagenicity. Specifically, benzofuroxan system seems to be responsible for the mutagenicity of certain agents against Chagas disease and certain anti-inflammatory agents. However other N-oxides, such as furoxans with anti-inflammatory and anti-atherosclerosis activities, seem to lack mutagenicity. In other cases, such as quinoxaline dioxides with anti-parasitic activity, mutagenicity shows to be substituent dependent. Applying CODES neural network two models were defined, one without metabolism and other with metabolism. These models predict the mutagenicity with and without metabolism in an excellent manner.