EEG signals are highly correlated, both in space (electrodes) and time (samples). Feature extraction indicates a wide range of signal processing methods to represent EEG signals by a small set of non-redundant values, named features. EEG feature extraction is usually described as the second signal processing step of EEG-based BCI design: it follows EEG pre-processing and is followed by classification, which will identify the class to which the EEG signal belongs, e.g., the user mental state. This chapter is an introductory overview on feature extraction methods. We will show how to extract relevant and robust spectral, spatial and temporal features from noisy EEG signals in order to classify them more efficiently. We will cover classical feature extraction methods such as band-power and time-embedded features, as well as spatial filtering methods such as the Common Spatial Patterns (CSP) and xDAWN. We will also briefly introduce a recent and promising alternative approach based on Riemannian information geometry. The overall objective of this chapter is to provide the reader with practical knowledge about how to extract features from EEG signals for BCI purposes, as well as to stress the key points of each approach.