Abstract. ElectroEncephaloGraphy signals have been studied in relation to emotion even prior to the establishment of Affective Computing as a research area. Technological advancements in the sensor and network communication technology allowed EEG collection during interaction with low obtrusiveness levels as opposed to earlier work which classified physiological signals as the most obtrusive modality in affective analysis. The current article provides a critical survey of research work dealing with broadly affective analysis of EEG signals collected during natural or naturalistic interaction. It focuses on sensors that allow such natural interaction (namely NeuroSky and Emotiv), related technological features and affective aspects of applications in several application domains. These aspects include emotion representation approach, induction method and stimuli and annotation chosen for the application. Additionally, machine learning issues related to affective analysis (such as incorporation of multiple modalities and related issues, feature selection for dimensionality reduction and classification architectures) are revised. Finally, future directions of EEG incorporation in affective and natural interaction are discussed.