This systematic review examines the application of deep learning (DL) to electroencephalography (EEG) signals, focusing on 156 papers published between January 2010 and July 2018. The review covers various application domains such as epilepsy, sleep, brain-computer interfacing, and cognitive and affective monitoring. Key findings include: 1. **Data Usage**: The amount of EEG data used varies from less than ten minutes to thousands of hours, with the number of samples seen during training ranging from a few dozen to several million. 2. **Public Data**: Over half of the studies used publicly available data, indicating a growing trend towards open data sharing. 3. **Methodology Shift**: There has been a shift from intra-subject to inter-subject approaches, reflecting a move towards more generalizable models. 4. **DL Architectures**: Convolutional neural networks (CNNs) were the most commonly used architecture, followed by recurrent neural networks (RNNs). CNNs were used in about 40% of the studies, while RNNs were used in 14%, often with 3 to 10 layers. 5. **Preprocessing and Feature Extraction**: Almost half of the studies trained models on raw or preprocessed EEG time series, highlighting the potential for automatic feature learning. 6. **Performance**: The median gain in accuracy of DL approaches over traditional baselines was 5.4%, but many studies suffer from poor reproducibility due to the unavailability of data and code. 7. **Future Recommendations**: The review provides recommendations for future research, including guidelines for data collection, preprocessing, and model training to enhance reproducibility and efficiency. The review aims to inform future research and contribute to the development of more effective and flexible EEG processing methods using deep learning.This systematic review examines the application of deep learning (DL) to electroencephalography (EEG) signals, focusing on 156 papers published between January 2010 and July 2018. The review covers various application domains such as epilepsy, sleep, brain-computer interfacing, and cognitive and affective monitoring. Key findings include: 1. **Data Usage**: The amount of EEG data used varies from less than ten minutes to thousands of hours, with the number of samples seen during training ranging from a few dozen to several million. 2. **Public Data**: Over half of the studies used publicly available data, indicating a growing trend towards open data sharing. 3. **Methodology Shift**: There has been a shift from intra-subject to inter-subject approaches, reflecting a move towards more generalizable models. 4. **DL Architectures**: Convolutional neural networks (CNNs) were the most commonly used architecture, followed by recurrent neural networks (RNNs). CNNs were used in about 40% of the studies, while RNNs were used in 14%, often with 3 to 10 layers. 5. **Preprocessing and Feature Extraction**: Almost half of the studies trained models on raw or preprocessed EEG time series, highlighting the potential for automatic feature learning. 6. **Performance**: The median gain in accuracy of DL approaches over traditional baselines was 5.4%, but many studies suffer from poor reproducibility due to the unavailability of data and code. 7. **Future Recommendations**: The review provides recommendations for future research, including guidelines for data collection, preprocessing, and model training to enhance reproducibility and efficiency. The review aims to inform future research and contribute to the development of more effective and flexible EEG processing methods using deep learning.