Time history analysis of infrastructures like dams, bridges and nuclear power plants is one of the fundamental parts of their design process. But there are not sufficient and suitable site specific earthquake records to do such time history analysis; therefore, generation of artificial accelerograms is required for conducting research works in this area.  Using time series analysis, wavelet transforms, artificial neural networks and genetic algorithm, a new method is introduced to produce artificial accelerograms compatible with response spectra for the specified site condition. In the proposed method, first, some recorded accelerograms are selected based on the soil condition at the recording station. The soils in these stations are divided into two groups of soil and rock according to their measured shear wave velocity. These accelerograms are then analyzed using wavelet transform. Next, artificial neural networks ability to produce reverse signal from response spectra is used to produce wavelet coefficients. Furthermore, a genetic algorithm is employed to optimize the network weight and bias matrices by searching in a wide range of values and prevent neural network convergence on local optima. At the end site specific accelerograms are produced. In this paper a number of recorded accelerograms in Iran are employed to test the neural network performances and to demonstrate the effectiveness of the method. It is shown that using synthetic time series analysis, genetic algorithm, neural network and wavelet transform will increase the capabilities of the algorithm and improve its speed and accuracy in generating accelerograms compatible with site specific response spectra for different site conditions.