Rainfall prediction is one of the major concerns in developing countries like Ethiopia. Ethiopian economy dependent on agriculture, so rainfall prediction is needed. Prediction in meteorology can assist in decision-making processes carried out by organizations re-sponsible for the prevention of disasters, drought, flood, risk management, water resource managements, hydropower etc. Several techniques have been done to predict rainfall based on statistical analysis and machine learning, but the papers that was done before has limitation on the model performance. This work implement LSTM based Recurrent Neural Network (RNN) to monthly rainfall prediction for Bahir Dar city. In this study we have used 34 years meteorological dataset that taken from Bahir Dar city Meteorology Branch. We have Pre-processed our data first missing value handling using mean/median technique then we have normalized our dataset using scaling techniques range between [0, 1]. The dataset split into training, validating and testing dataset. Mean squared error taken as a metrics and calculate the error based on actual and predicted data. Stochastic Gradient Descent (SGD) algorithm implemented on this study for error optimization. The parameters considered for the evaluation of the performance of rainfall prediction model are number of epochs, batch size, and learning rate of the network. The monthly rainfall predictions obtained after training and testing are compared with actual data to ensure the performance of the neural model. As results of this study the model performance is 94.15% and mean squared error is 5.85%. Recurrent neural network model is most suita-ble among the artificial networks for the rainfall prediction.