Accurate streamflow forecasting can help minimizing the negative impacts of hydrological events such as floods and droughts. To address this challenge, we explore artificial neural network models (ANNs) for streamflow forecasting. These models, which have been proven successful in other fields, may offer improved accuracy and efficiency compared to traditional conceptually based forecasting approaches. The goal of this study is to compare the performance of a state-of-the-art conceptual rainfall-runoff (hydrological) model with an artificial neural network (ANN) model for streamflow forecasting. As a test area, we use the Severn catchment in the United Kingdom. The adopted ANN model has a long short-term memory (LSTM) architecture with two hidden layers, each with 256 neurons. The model is trained on a 2-year dataset (2016-2017) and validated on a 3-year dataset (from 2018 to 2020), and also on a 14-year dataset (2004-2017) for training, validated over the same period as previously. 2019 was chosen in the validated period for being a particularly wet year relevant to assess the model performance in extreme hydrological conditions. The study focuses on daily and hourly predictions. The Severn River is the longest in Great Britain and runs from its source in the Welsh hills to the Bristol channel. The basin is mainly rural, with some urban settlements like Worcester, Tewkesbury and Evesham. The study area focuses on the area on Saxons Lode. The data of our study is provided by ERA5, which is a dataset that provides hourly estimates of various climate variables such as precipitation, temperature, mean surface downward short-wave radiation flux, snow-depth, atmospheric and dewpoint temperature. The data covers the entire Earth on a 0.25-degree grid and has a detailed resolution of the atmosphere using 137 levels from the surface up to 80km high. Our model is based on a LSTM, that is a type of Recurrent Neural Network (RNN) architecture, specifically designed to handle sequential data with long-term dependencies. LSTMs can retain information for longer periods of time by using gates to control the flow of information through the network. These gates can decide which piece of information to keep and which to discard, enabling LSTMs to remember important information from earlier in the sequence while also being able to forget irrelevant information. LSTMs are commonly used in natural language processing, speech recognition, and time-series forecasting tasks. As can be seen in fig.2, a scheme of a two-layer LSTM model used for our study. To carry out this study, the conceptual hydrological model called Superflex is used as a benchmark. Superflex is a flexible framework for conceptual hydrological modeling who aims to generalize and systematize the field of conceptual models and provide a robust platform for understanding and modeling hydrological systems at the catchment scale. It allows hydrologists to use a combination of generic components to hypothesize, build, and test different model structures, which is useful due to the limitations in process understanding and data availability at this scale [3]. Both models are first evaluated using the Nash-Sutcliffe Efficiency (NSE) score. To enable a meaningful and fair comparison, both models share the same inputs (i.e., meteorological forcings: total precipitation, daily maximum and minimum temperatures, daylight duration, mean surface downward short-wave radiation flux, and vapor pressure) and share the same training and validation period. The ANN model was implemented using the Neuralhydrology library developed by F. Kratzert. In our study, for the 2-years training period, we found that LSTM model can provide more accurate one-day forecasts than the hydrological model Superflex. For the daily predictions, the average NSE score using the LSTM model is 0.81 for validation period (With an average NSE score of 0.96 for training period), which is higher than the NSE score of 0.76 achieved by the Superflex model (with a score of 0.70 for training period). During the 14-year training period, we obtained an average NSE score of 0.88 using the LSTM model, which had an average NSE score of 0.97 during the training period. The Superflex model had an NSE score of 0.72 (and 0.70 during the training period). We have conducted the same experiments for the hourly time step, and we obtained similar results and conclusions to those of the daily time step. These results were obtained without adjusting the hyperparameters and by training the model only on data from the Severn watershed. The ANN model has demonstrated promising results compared to a state-of-the-art conceptual hydrological model in our studies. We will further compare both models using different training dataset periods, and different catchments. These additional tests will provide more information on the capabilities of the LSTM model and help to confirm its effectiveness.