The detection of a stimulus in the environment largely depends on its salience (e.g. an intense flash of light is easier to detect than a faint light) but the state of the brain has also a strong influence on the perception and response to stimuli. Sleep is a typical and extreme example of such influence. It is thought that the sensory isolation that operates during sleep serves vital functions and that the processing of irrelevant stimuli is impaired during sleep. Yet, what makes a stimulus relevant for the sleeper remains to be elucidated. Sleep is a composite state divided in several sleep stages with distinct behavioral, physiological, electrophysiological, and phenomenal manifestations that reflect the action of different functions. The perceptual dimensions that make a stimulus relevant should therefore depends on the sleep functions that operate and therefore, on sleep stages. In this thesis, I investigated the relation between vigilance states and which perceptual dimensions are relevant to the sleeping brain by recording the cerebral response during sleep to stimuli of different natures. First, I recorded the brain response to interoceptive signals (i.e. heartbeat) in sleep stages with distinct degrees of sensory disconnection. I found a consistent change of brain response to visceral signals between wakefulness and sleep and between sleep states of high and low sensory disconnection. Thus, although interoceptive signals convey to the brain crucial information about the body and would be expected to be constantly relevant, their processing changes with sleep stages. Then, I investigated the brain response to intrinsically aversive stimuli (screams) during sleep. I found that screams evoked more sleep oscillations than controlled “neutral” vocalizations. This result suggests that intrinsically alarming signals conserve their relevant status during sleep. Next, I operated a conditioning procedure on participants to associate an auditory cue with humorous pictures and another auditory cue with neutral pictures. These participants were medically resistant epileptic patients implanted with intracranial electroencephalography for pre-surgical evaluation. I benefited from rare intracranial recordings to investigate the response of specific brain regions to the two conditioned auditory cues during sleep. I found that the orbitofrontal cortex, a brain region crucially implicated in the evaluation of the affective properties of stimuli evoked more sleep oscillations in response to the auditory cues paired with emotional items. This result suggests that emotion-related brain regions react to relevant stimuli during sleep. Finally, I recorded participants with electroencephalography and polysomnography while they were stimulated with odors overnight. I recorded a response of the brain to odors in wakefulness and sleep which supports the detection of odors by the sleeping brain. In addition, odors modified the sleep architecture: they reduced the sleep latency but fragmented deep sleep. This new knowledge informs about which stimuli can fragment human sleep, can influence sleep state, or can evoke complex sensory representations. In the long term, they could serve to inform public health policies and design overnight stimulation to treat disorders including insomnia.