 This paper proposes a new data-driven approach to rainfall runoff modeling by using a long short-term memory, LSTM, network, a type of recurrent neural network. This approach was tested on 241 catchments of the freely available camels dataset and compared to the Sacramento Soil Moisture Accounting Model, SSSMA, coupled with the Snow 17 Snow routine. Additionally, the potential of the LSTM as a regional hydrological model was demonstrated by transferring knowledge gained from regional scale to individual catchments. The results showed that the LSTM outperformed the SSSMA in Thinspe, plus in Thinspe, Snow 17 model, indicating the potential of the LSTM for hydrological modeling applications. This article was authored by F. Krotzert, F. Krotzert, D. Klotz, and others.