The laminar flamelet concept for prediction of temperature and species mass fractions in non-premixed turbulent flames using appropriate kinetic mechanism, traort and thermodynamic properties has been investigated. In a first step, temperature and species concentrations for a flame have been calculated from the solutions of counterflow diffusion flames considering differential diffusion effects and related to mixture fraction and scalar dissipation rate which accounts for flow field effect on flamelets structure. In a second step numerical integration is accomplished for different values of mean mixture fraction, its variance and dissipation rate to obtain favre averaged values of species concentration and temperature and flamelet library is created. Finally two Multi Layer Perceptron(MLP) networks are built over this library and trained by means of Error Back Propagation (EBP) algorithm where mean and variance of mixture fraction and dissipation rate are inputs and species mean mass fraction and temperature are outputs of these two networks. The ANN weights and biases are applied to a CFD flow solver code and results are compared with experimental and other numerical methods. ANN yields accurate and acceptable results which are calculated much faster in comparison to when averaging is implemented in a CFD code. Species mass fraction and temperature predicted by flamelet model are qualitatively identical to those of experimental measurements but some discrepancies are observed which can be described by non-unity Lewis number assumption. Unity Lewis number calculations are in better agreement with experimental ones specially for CO 2 mass fraction and at higher axial positions.