In this study daily mean river flow estimations based on the antecedent rainfall and/or river flow values are carried out using two different ANN algorithms. Estimations obtained by the conventional feed forward back propagation (FFBP) and the new generalized regression neural network (GRNN) methods using different network architectures are compared with conventional statistical and stochastic model results. GRNN is used for estimation of continuous variables, as in standard regression techniques. Generalized regression neural network does not require an iterative training procedure as in back propagation method. In the applications presented in this study the daily hydro meteorologic data belonging to Juniata River Basin in Pennslyvania state of USA are employed. The FFBP results are found to be superior to GRNN and conventional methods. However it was shown that FFBP method performances are very sensitive to the randomly assigned initial weight values. This problem, however, was not faced in GRNN simulations. The local minima problem in feed forward back propagation algorithm applications was not faced in GRNN simulations. In the study it was also seen that the analysis of the cross-correlations and autocorrelations provide useful information about the input node number of the ANN network.