Hydrological processes, as complex systems, and their modeling procedures, either physically-based or numerically, are difficult tasks due to their inter-relational effects of parameters and variables. One of the important of these processes, related to the management of water resources, is the derivation of reservoir operation rules in or to determine the optimal reservoir releases as the main sources of supply water. The accuracy and reliability of the water supply determination is based on good estimate and reliable forecasting of reservoir inflow and the duration of projection. In addition, the inherent uncertainty in runoff prediction makes the modeling procedures very sensitive to both data and type of models. This research focuses on a use of a newly developed integrated model, namely Adaptive Neuro-Fuzzy Inference System (ANFIS), to investigate the potentials of neuro-fuzzy systems in modeling runoff time series into Kardeh reservoir in Khorasan province and to access its performance relative to artificial neural networks (A). This method has demonstrated very satisfactory performances in applying to different hydrological process analysis. However, when the method encounters large number of parameters, including input and output variables, it produces many rules and faces curse of dimensionality during computational steps. Employing different fuzzy clustering preprocessing procedures classifies the input data into different independent groups and selects the most effective set of input data in order to reduce the dimension of dataset and enhance the efficiency of algorithm by formation of fewer numbers of fuzzy laws. Subtractive Clusrering (SUBCLUST) and Fuzzy C-Means (FCM) are two methods of fuzzy clustering employed in this research. In accordance with the main objective of this research, two daily and monthly models have been developed using Adaptive Neuro-Fuzzy Inference System with a preprocessing step of clustering the input dataset. daily models for short-term decisions such as reducing flood losses and monthly models for medium-term decisions such as operation of reservoirs. Generally, the results showed that the Adaptive Neuro-Fuzzy Inference System forecasted inflow series preserve more characteristics of actual inflow data in comparison with artificial neural networks. Also, Fuzzy C-Means has demonstrated a relatively better result compared with Subtractive Clusrering algorithm as far as the classification of the data was concerned. The last phase of this research was to utilize the forecasted inflow for determining the operational rules of the Kardeh reservoir. Optimal operational rule-curve based on the result of Harmony Search algorithm were derived and compared with traditional simulated rule-curves. The results indicated more uniform monthly supply of water demand during a year relative to previously simulated outcomes.