The paper presents activities performed in the frame of project Clean Sky II DECK in Large Passenger Aircraft Platform 3 and related to the detection of sleep by processing the multimodal data acquired from subjects using various kinds of sensors. The data acquisition was carried out in a cockpit simulator, with each subject participating in two sessions, resulting in more than 40 hours of recorded data. The polysomnography (PSG) data were scored online by a medical expert according to the guidelines of the American Academy of Sleep Medicine (AASM). Several feature extraction and selection techniques were employed to identify the most relevant characteristics for discriminating between sleep and alert states in aviation context. The experimental results suggest that the proposed data-driven paradigm for sleep detection achieves high accuracy and shows high robustness when combining the most discriminative features extracted from the considered modalities.