Access to sufficient and confident hydrometric data is necessary for water resources management. Most of the Iran's hydrometric stations do not have sufficient data. The method of producing synthetic data should use probability concepts and retains main characteristics of the data, too. In this research, synthetic hydrometric data are generated by the monthly and annual Markov chain method at the Telezang station in the upstream of the Dez River. Using the discharge of the driest day and the wettest day of each month and the generated monthly hydrometric data, the probable highest and lowest daily discharge for each month was calculated. At the end, artificial neural network was trained with a number of observed and generated hydrometric data. The results of artificial neural network were compared with a number of observed hydrometric data which were not used in training of the network. The training of artificial neural network (ANN) with the generated hydrometric data can improve results of network. For more improvement of the results of network, genetic algorithm (GA) is used in its training and optimizing its parameters. The GA method can reduce the MSE (mean of square error) by 97% that of ANN.