The decomposition behavior of energetic materials is very important for the safety problems concerning their production, transportation, use and storage, because molecular decomposition is intimately connected to their explosive properties. Nitroaromatic compounds, particularly nitrobenzene derivatives, are often considered as prototypical energetic molecules, and some of them are commonly used as high explosives. Quantitative structure-activity relationship (QSAR) represents a potential tool for predicting the thermal stability properties of energetic materials. But it is reported that constructing general reliable models to predict their stability and their potential explosive properties is a very difficult task. In this work, we make our efforts to investigate the relationship between the molecular structures and corresponding thermal stabilities of 77 nitrobenzene derivatives with various substituent functional groups (in ortho, meta and/or para positions). The proposed best MLR model, developed by the new software QSARINS, based on Genetic Algorithm for variable selection and with various validation tools, is robust, stable and predictive with R(2) of 0.86, QLOO (2) of 0.79 and CCC of 0.90. The results indicated that, though difficult, it is possible to build predictive, externally validated QSAR models to estimate the thermal stability of nitroaromatic compounds.