This paper proposes an irregularity detecting monitoring system of gravity dam related with a monitoring data of leakage by using the multi-patterns differentiating neural network system. The proposed system predicts the leakage amount with respect to the water level of dam, precipitation, temperature of water, air and leakage water every day. The difference between the prediction and the observation of leakage notices an irregularity of the dam. The real observation data used herein contains an irregularity induced by giving an artificial noise. The neural network system detects this irregularity much better than the multi-regression system. The proposed system detects it better than conventional neural network system, because the observed data includes multi-patterns relationship between input and output data.