The predictive capability of computational fluid dynamics (CFD) fire models is highly dependent on the accuracy with which the source term due to fuel pyrolysis can be determined. The pyrolysis rate is a key parameter controlling fire behavior, which in turn drives the heat feedback from the flame to the fuel surface. The main objectives of the present study were twofold. First, an in-depth pyrolysis model of a semi-transparent solid fuel (here, clear poly-methyl-methacrylate or PMMA) with in-depth radiation and a moving gas/solid interface was coupled with a CFD code which included turbulence, combustion and radiation for the gas phase. Second, experiments were conducted in order to validate coupled model results. A combined genetic algorithm/pyrolysis model was used with Cone Calorimeter data from a non flaming pyrolysis experiment to estimate a unique set of kinetic parameters for PMMA pyrolysis. Flaming experiments were conducted on square slabs of PMMA with side dimensions of 10, 20 and 40 cm. From data collected at the center of the slabs, it was found that i) for any sample size, the experimental regression rate remains almost constant with time, with average values of 5.8, 8.6 and 10.9 mu m s(-1) for the PMMA slabs with side lengths of 10, 20 and 40 cm respectively, and ii) although the radiative and total heat transfers increase significantly with sample size, the radiative contribution to the total heat flux remains almost constant (similar to 80%). Coupled model results show a fairly good agreement with the literature and with current measurements of the heat fluxes, gas temperature and regressing surface rate at the center of the slabs. Predicted flame heights based on a threshold temperature criterion were found to be close to those deduced from the correlation of Heskestad. However, discrepancies between predicted and measured total pyrolysis rates are observed, which result from the underestimation of the flame heat feedback at the edges of the slab, as confirmed by the comparison between predicted and observed topography of burned samples. (C) 2016 The Combustion Institute. Published by Elsevier Inc. All rights reserved.