This study aims to improve the description of gas phase combustion in physical models of forest fire spread. The models operating at field scale liken the degradation gases of forest fuels to carbon monoxide burning in air, whatever the vegetation species, The first part of the Study was devoted to determining whether the degradation gases have to be considered accurately in forest fire modeling. A laboratory experimental apparatus was designed to Study the influence of the degradation gases oil the proprieties of laminar flames from crushed forest fuels. Thanks to these experiments, the role of the degradation gases in gas phase combustion was highlighted. The second part was dedicated to improving the combustion models of degradation gases used in the modeling of forest fire behavior. Using numerical methods, the equations for the conservation of mass, momentum, energy, and chemical species were solved, as well as the radiative transfer equation for a laminar flame. Skeletal and global combustion mechanisms, including the main degradation gases released by forest fuels, were tested. The numerical predictions were evaluated by comparisons with measured temperature and flux distributions. The computational time was considered as a criterion of comparison between the combustion mechanisms, as well. The skeletal mechanisms provide results Close to those from experiments. However, they require excessively long Computational times because of the number of elementary reactions. Comparisons between observed and predicted temperatures indicated that the global mechanism considering only carbon monoxide significantly underestimated the temperature in the fire plume. In contrast, the predictions of the global mechanism including both carbon monoxide and methane with incomplete oxidation of methane matched the experimental data. This global mechanism is reliable and time-saving. meeting the requirements for use in physical models of forest fires. (C) 2009 The Combustion Institute. Published by Elsevier Inc. All rights reserved.