An artificial neural network model with three-layer structure was applied to estimate chlorophyll a concentration in a lake located in Tottori Prefecture, Japan. First, input variables, which resulted in high calibration accuracy, were searched to examine the optimal network structure. The calibration accuracy was highest when input variables were set to TN, TP, DO, water temperature, solar radiation, air tempera- ture, wind velocity, and Wedderburn number. This result means that the model incorporated the rela- tionship between chlorophyll a concentration and the meteorological, hydraulic, and aquatic factors into the network structure. The adaptability of the estimation of chlorophyll a concentration was examined. As a result, chloro- phyll a concentration could not be sufficiently estimated. To improve estimation accuracy, the network structure was reconstructed by considering the time history of the variation of the meteorological and water quality data for the previous 24 hours and incorporating such data into the input variables. The result showed that the estimation accuracy was remarkably improved.