This systematic review analyzes 156 papers applying deep learning (DL) to electroencephalography (EEG) between 2010 and 2018. The study explores trends and highlights promising approaches in various domains, including epilepsy, sleep, brain-computer interfacing, and cognitive/affective monitoring. The analysis reveals that EEG data used in studies varies from less than ten minutes to thousands of hours, with training samples ranging from dozens to millions. Over half the studies used publicly available data, and there has been a shift from intra-subject to inter-subject approaches. Convolutional neural networks (CNNs) were used in 40% of studies, while recurrent neural networks (RNNs) in 14%. Almost half of the studies trained on raw or preprocessed EEG data, and the median accuracy gain of DL over traditional methods was 5.4%. However, many studies suffer from poor reproducibility due to lack of data and code availability. The review also discusses challenges in EEG processing, such as low signal-to-noise ratio, non-stationarity, and inter-subject variability, and highlights how DL can improve generalization and flexibility. The study provides recommendations for future research and makes a summary table of DL and EEG papers available for contribution. Key findings include the use of CNNs and RNNs, the importance of data preprocessing, and the potential of DL in improving EEG processing. The review emphasizes the need for reproducible research and highlights the growing interest in DL-EEG applications.This systematic review analyzes 156 papers applying deep learning (DL) to electroencephalography (EEG) between 2010 and 2018. The study explores trends and highlights promising approaches in various domains, including epilepsy, sleep, brain-computer interfacing, and cognitive/affective monitoring. The analysis reveals that EEG data used in studies varies from less than ten minutes to thousands of hours, with training samples ranging from dozens to millions. Over half the studies used publicly available data, and there has been a shift from intra-subject to inter-subject approaches. Convolutional neural networks (CNNs) were used in 40% of studies, while recurrent neural networks (RNNs) in 14%. Almost half of the studies trained on raw or preprocessed EEG data, and the median accuracy gain of DL over traditional methods was 5.4%. However, many studies suffer from poor reproducibility due to lack of data and code availability. The review also discusses challenges in EEG processing, such as low signal-to-noise ratio, non-stationarity, and inter-subject variability, and highlights how DL can improve generalization and flexibility. The study provides recommendations for future research and makes a summary table of DL and EEG papers available for contribution. Key findings include the use of CNNs and RNNs, the importance of data preprocessing, and the potential of DL in improving EEG processing. The review emphasizes the need for reproducible research and highlights the growing interest in DL-EEG applications.