In this research, a combined hybrid artificial neural networks and genetic algorithms method was used to invert a four-layer sounding model including eight different curve types. For this purpose, 2000 sounding data was synthetically generated using Resix-IP software. Then, different types of curves were classified applying a feed forward artificial neural network with back propagation algorithm based on a trial and error process of training data. The network of classification included 22 neurons in input layer, 33 neurons in hidden layer and 8 neurons in output layer. Subsequently, the inversion of four-layer model of geoelectrical sounding data was conducted by genetic algorithms. The obtained results of resistivity values showed a very good agreement between the outputs of genetic algorithms and test data. For instance, the strong correlation can be reflected by coefficients (0.99, 0.82, 0.83 and 0.97) and (0.99, 0.92, 0.93 and 0.97) of resistivity values for layers No.1 to No.4 in turn in curve types AA and AK. Besides, in all of the curves, thickness of the first layer was appropriately estimated by genetic algorithms method, while in these two curve types, the correlation coefficients in the second layer (0.81 and 0.88) and the third layer (0.79 and 0.71) show a relative capability of this method.