Abstarct: Water saturation is one of the most important parameters of hydrocarbon reservoirs. Archie formula is best known technique for calculating water saturation. This formula includes three parameters as cementation factor (m), saturation exponent (n) and tortuosity (a) that are called Archie’s parameters. A slight change of any of these three parameters can cause significant change in water saturation calculated value. Although numerous methods have been presented so far for estimation of these parameters, unfortunately no reliable and accurate method exists that can estimate these parameters in any circumstances. In this thesis, we used statistical and neural network methods to estimate the Archie’s parameters in three wells located in a carbonate field, and the capability of each of the methods has been evaluated. In this regard, first, we computed the Archie’s parameters using two known statistical techniques, namely conventional method three-dimensional (3-D) regression method, and then, a comparison has been made between the results of these two methods. In sum, the 3-D regression method has had better results than conventional method because of simultaneous computation of these parameters by this method. The Archie’s parameters does not follow a specific process and depend on many factors. The behaviour of any of these parameters or coefficients is complex and vague. Hence, it is reasonable to use artificial neural network (ANN) method to estimate these parameters. As a result, this method has been used to compute the cementation factor in each sample in the study area or field. In this research work, feed forward back propagation error networks were trained using two methods including early stopping and regularization methods. For this purpose, well logging data including CGR, SGR, RHOB, DT, ILD, NPHI and PEF data were used as the inputs to the networks, and the cementation factor was considered as the output. Levenberg-Marquardt algorithm and bayesian regularization function were used for the training the networks in the early stopping and regularization methods, respectively. Due to the low number of samples and the inputs to the networks, the expected accuracy and soundness in the obtained results from the ANN method were not achieved. However, the early stopping method contained better results than the regularization method. The best network obtained with the early stopping method, was a three-laxyer network, in which the mean of square error (MSE) of 0.01 and determination coefficient of r = 91.02% in the training step were obtained.