Computational methods to predict material properties from their molecular structure, down-selection, and designing new materials, were used for material research and development. The Quantitative Structure-Property Relationships (QSPR) method for quick screening of materials based on correlations derived from existing data for a family of materials could be especially useful if the correlations include molecular-structure parameters. In this article several QSPR models, in the form of mathematical equations were developed from a dataset of a wide variety of polymers to predict their flammability characteristicsheat release capacity, total heat release, and % char, using genetic function algorithms. The models were used to predict the flammability characteristics of a small test set of polymers for model validation; and additional predictions were made on an external dataset of about 20 polymers. The QSPR predictions on the external dataset were compared with the molar group contribution method of Lyon et al. (Lyon, Takemori, Safronava, Stoliarov, and Walters, Polymer, 50, 2608 (2009)). An attempt was made to correlate the molecular descriptors, which could be calculated from the molecular structure, to the flammability behavior. Good correlation was obtained between the polymer repeat unit structure and the flammability parameters. Based on the molecular descriptors in the QSPR equations, a molecular level understanding of the flammability characteristics was obtained. POLYM. ENG. SCI., 55:1553-1559, 2015. (c) 2015 Society of Plastics Engineers