The limitations of fossil fuel resources and environmental issues have provided extensive research on combustion science over the past decades, and the reduction of combustion pollutants in industrial burners has been one of the fundamental challenges for researchers. To monitor and control pollutants, proper prediction of the temperature field and combustion products is essential. On the other hand, most of the combustion processes occur in a totally turbulent environment, which is complicated by the interaction between chemical reactions and turbulence. Therefore, the simulation of these flows requires the selection of a suitable combustion model. In the present study, the laminar flamelet model is used for combustion modeling, because of its high accuracy in predictions and computational economy, and the model is used for turbulence modeling. This study is concerned with numerical solution of multi-fuel combustion. To this end, a new program called multi-mixture fraction model has been developed for the modeling of multi-fuel burners, and its applications in using flamelet model and flame-sheet model have been investigated. In order to ensure the correct predictions of the modeling, the bluff-body flame is simulated with one fuel stream, and piloted methane-air flame with two fuel streams. Simulation of flamelet model requires generation of a flamelet library, which is produced as preprocessing both in the physical space and in the mixture fraction space. In this thesis, the opposed-flow flame in physical space has been used for flamelet library generation. In the next step, the mean of these quantities was obtained by numerical integration for different mean mixture fractions, their variance and scalar dissipation rate, and then by training an artificial neural network and inserting its weight and bias coefficients in the solution code for conservation equations, the average mass fraction of species is foreseen. The simulation results of the bluff-body flame showed that the prediction of temperature and mass fraction of species by the neural network in the flamelet code yields good agreement with experimental data. The flame-sheet model does not offer a good prediction of species, especially pollutants, so that the mass fraction of CO is much less than its real value. Another result of the simulation of this flame is the reduction of computational time with the help of neural networks, in which the application of this network in the flamelet code reduced the computational time to about 40 percent over the implementation of flamelet fluent. The simulation results of piloted methane-air flame using the multi-mixture fraction model were compared with experimental data as well as predictions of the flamelet fluent model. The flamelet model based on a single-mixture fraction predicts the mass fraction of CO and OH species more than real value in most situations. But the use of two-mixture fraction model reduced the prediction error for this species by about 5 and 8 percent, respectively. Also, the use of two-mixture fraction model reduced the temperature prediction error by about 11 percent compared to the single-mixture fraction model. The results of this simulation showed that the standard mixture fraction framework is not suitable for the modeling of multi-fuel burners, and the development of mixture fraction based on this study has the ability to solve combustion problems with more than one fuel input. Keywords: Combustion, Laminar flamelet model, Artificial neural network, Multi-mixture fraction