Currently determining the coagulant dosing rates depends on Jar-test results and the operators experience in many cases. The nature of these practices makes it difficult to timely cope with the rapid fluctuation of raw water quality mainly since it takes relatively long time to obtain Jar-test results. For promptly predicting required coagulant doses in response to water quality change, a number of researchers have attempted to use multi-variable regression (MVR) approach. However, it has been known that the prediction capability of the MVR. approach is not satisfactory. Artificial neural network (ANN) is known as an excellent estimator of nonlinear relationship between the accumulated input and output numerical data. Using this nature of the ANN, this study has attempted to predict the optimal coagulant dosing rate with accuracy and in time. The ANN used in this study consists of an input, a hidden and an output layer. The input and output relationship of a neural network can be nonlinear and linear, and its characteristics are determined by the weights assigned to the connections between nodes in two adjacent layers. Changing these weights will change the input/output behavior of the network. Systematic ways of determining the weights of the network to achieve a desired input/output relationship are referred to as a training or learning algorithm. In this study, quasi-Newton algorithm is used as a training method which is known most efficient for the unconstrained optimization. To train the ANN and deduce the MVR equation, a set of 142 data chosen from the 2-year operation of a water treatment plant was used. Another set of 72 data not used in training was also used to check the prediction capability of the trained ANN and MVR equation. Mean square root error(MSRE) has been used as a quantitative indicator of prediction capability. For the training data set and the raw data set, MSREs of the MVR equation are, respectively, 0.0143 and 0.0193 while those of the ANN 0.0058 and 0.0092, respectively. These results indicate that the ANN reduced the prediction error for the training data by about 59% and for the raw data by about 52%. Therefore, it can be said that the prediction capability of ANN for raw data is enhanced about twice as much as that of MVR. As the advancement of on-line monitoring techniques enables ANN to continuously update the weights periodically, its prediction capability can be also continuously enhanced. However, it is noted that since ANN can not reveal the direct mechanistic relationship between water quality parameters and required coagulant doses, the training data must be prepared correctly in advance to reliably use the ANNs prediction. Further research is necessary for improving correctness of the training data.