Abstarct: One of the most important tasks in quantitative reservoir characterization is water saturation prediction. Water saturation is a parameter which helps evaluating the volume of hydrocarbon in reservoirs. To determine this parameter, many approaches such as cores and logs analysis were introduced thorough last decade. Treating the problem of dependency on core analysis in previous works, other scientists proposed using seismic data and arrived at improved models of water saturation estimation. This thesis aims at improving the methods of determination of water saturation in carbonate hydrocarbon reservoirs. One of the Iranian carbonate reservoirs in the south-western part of Iran was used for developing and testing our procedures. Monitoring of velocity values from sonic logs has exhibited inversion in this reservoir. We attribute this inversion to the change in pore sizes. To obtain real values of dry rock bulk modulus as an indicator of pore sizes, assuming an identifiable model, we devised a genetic algorithm to optimize the Gassmann velocity equation. Consequently, a proposal for modification of the Gassmann velocity equation is presented by introducing a new coefficient representing the effects of pore sizes. In the next step, 81 different synthetic models of porosity and pore size were constructed using modified Gassmann velocity equation and Seismic Unix forward modeling package. Extracting 43 attributes and performing sensitivity analysis on these attributes showed that the best attributes correlate with the values of porosity and pore size are Envelope Weighted Phase and Frequency for porosity and Instantaneous Amplitude and Asymmetry for pore size, respectively. Two modeling methods of reservoir parameter were used to determine the unknown nonlinear relationships between proper attributes and the values of porosity and pore size. A network of artificial neurons and a machine of support vectors were trained using the outputs of synthetic models that were assigned for training of these two machines. Finally, both of the abovementioned machines were used to relate the values of porosity, pore size and P-wave velocity with the values of water saturation. Considering the RMS error values of 0.04, 0.09 and 0.06 for each prediction, the proposed SVM method is able to predict the values of porosity, pore size and water saturation precisely.