Abstarct: Porosity is one of the most important properties for comprehensive studies of hydrocarbon reservoirs. For determination of porosity in a rock, that is the ratio of volume of voids to the total volume of the rock, there are two conventional methods: In the first method, direct measurement of porosity is carried out by testing drilling cores. In the second method, porosity is determined indirectly using well logging data and relevant mathematical relations or equations. There are some limitations and difficulties for determination of porosity using the above both methods. Using artificial neural networks (ANNs) method for this purpose can reduce these difficulties remarkably, and in the meantime, contains acceptable results. Multiple networks systems, which are also called committee machine, are proposed for increasing accuracy and decreasing errors due to usual artificial neural networks. In this research, for the first time in Iran, ensemble combination of networks, as a kind of committee machine, is used in order to determine effective porosity of Kangan gas reservoir rock in giant Southern Pars hydrocarbon field. Well data (4 wells) in the depth range related to Kangan formation were used. Acoustic, density, gamma ray and neutron porosity well logging data were used as input data for the networks, and effective porosity was considered as the output of the networks. The research was carried out in three parts. In the first part, back-propagation (BP) trained networks using overtraining and combination of the networks were employed, and as a result, the simple averaging, optimal linear combination (OLC) and non-linear combination with another network, decreased the mean of squares of errors (MSE) for estimation of test patterns, compared to the result of best single network, to 3.9 %, 4.9 % and 19 %, respectively. In second part, BP trained networks using regularization method were prepared and then the results were combined. In this part, the MSE for estimation of test patterns, compared to the result of best single network, was reduced to 3.4 %, 4 % and 21.6 % using simple averaging, OLC and non linear combination with another network, respectively. In third part, BP trained networks using early stopping and combination of the networks were employed, and as a result, the methods of simple averaging, OLC and non-linear combination with another network, decreased the MSE for estimation of test patterns, compared to the result of best single network, to 2.3 %, 2.8 % and 2 %, respectively.