Increase of oil consumption and price of the crude oil in the world, and declining oil supplies in recent years, has caused an increasing attention to heavy oil and extra-heavy oil production methods. As far as substantial parts of such reservoirs are fractured and fractured reservoirs are not easily exploited by using conventional technologies and requires more energy, time and high cost, so, more recognition of oil recovery methods from fractured reservoirs and finding details of each methods, especially the correct understanding of effective oil production mechanisms, are effective in selection of suitable and economical methods.So, in this study, three dimensional modeling of steam injection in a fractured heavy-oil carbonate reservoir was performed. First, steam injection process was modeled in a non-fractured reservoir, then, the model accuracy was assessed.This model showed good results compared with experimental results model (Willman and Shutler).This model considered both capillary and gravity effects in the energy equation and a variable porosity. The results showed that steam injection in a non-fractured reservoir increases the oil recovery to 70 %. In the next step, steam injection process was modeled in a fractured reservoir. The amount of oil recovery in this reservoir was estimated about 30 %.The sensitivity analysis for identification and evaluation of steam injection process in both non-fractured and fractured reservoirs was performed. The results indicated that some parameters such as: injection pressure, steam quality, pre-heating, permeability and location of fracture affect on the oil recovery. Oil recovery from carbonate fractured reservoirs is less compared with non-fractured reservoirs, and breakthrough occurs sooner. Steam injection, pre-heating and cyclic steam injection into Kuh-e-mond reservoir has been studied, too. The results showed that pre-heating and cyclic steam injection increase oil recovery. Finally, our modeling results were compared with the results of the CMG simulator. The amount of oil recovery obtained from our modeling has an error of 1.67 % compared with the CMG result in a horizontal fractured reservoir. In the next part, steam distillation mechanism during steam injection was studied and the yield of steam distillation as an effective oil production mechanism was predicted by the artificial neural network model. Multi-Layer-Perceptron was chosen as our neural network model. Different one-hidden layer and two hidden layer networks built by our neural network model and examined by input parameters. We used Supervised training algorithm so the desired responses (steam distillation yield) entered as outputs of models. After building the main structure of neural networks, step of training and testing begun, finally in cross-validation step, the simulation results compared with experimental results. In this comparison the numbers of neurons in hidden layers changed, for example four types of One-hidden layer and two hidden layer neural networks by different neurons number(3, 4, 5, 6) built and were investigated. The results showed one hidden layer neural network had less average relative error (ARE %) compared with some previous models. The average relative error of this model was obtained about 2%. Keywords Enhanced oil recovery, steam injection, modeling, fractured reservoir, neural network, multi-Layer perceptron, steam distillation mechanism.