Abstarct: Water saturation is one of the most important and applied parameters in study of hydrocarbon reservoirs, which defined as percentage of empty spaces filled by water. In this study, we've used Committee neural networks to estimate the water saturation parameter in the study. To this end we got benefit the data obtained from 5 wells locating in a southwestern field of Iran. The values of numerical sonic log, density, Gamma ray, resistivity, long normal resistivity, and effective porosity (which obtained by various logs and cores) are considered as the input, whereas water saturation is considered as the output. This method is more drastic and precise than using a single network method. The elements of the Committee neural networks are multilxayer perceptron networks. The best structure of them was selected by trial and error method regarding the least test MSE. 59 networks with various structures have been trained and tested. The training for these 59 networks has been done by over-training, regularization and early stopping. Among the 59 networks, 7 networks with least MSE were selected for construction of ensemble combinations and 120 possible ensemble i.e. 21 two-network combination, 35 three-network combination, 35 four-network combination, 21 five-network combination, 7 six-network combination, 1 seven-network combination was constructed. Weight coefficients of the linear combinations have been obtained by 2 methods: Genetic Algorithm and Simple Averaging. Then the results of the combinations have been compared with respect to each other. For over-training method, the single laxyer network with the structure of (6-10-1) has been known as the best. When the weight coefficients of linear combination of networks (6-10-1), (6-6-5-1), (6-13-1), (6-11-1) and (6-4-9-1) is obtained by using the genetic Algorithm, the most decrease in test MSE is observed in combination of these networks. In regularization method, a network with the structure (6-10-2-1) has the least MSE in generalization stage. When the weight coefficients of linear combination of networks (6-10-2-1), (6-6-5-1), (6-5-6-1), (6-4-10-1) and (6-6-6-1) is obtained by using the genetic Algorithm, the most decrease in test MSE is observed in combination of these networks. In early stopping method, the network with structure (6-8-8-1) When the weight coefficients of linear combination of networks (6-8-8-1), (6-8-9-1), (6-15-1), (6-6-9-1) and (6-7-10-1) is got by using the genetic Algorithm, the most decrease in test MSE is observed in combination of these networks.