Although transit spectroscopy is a powerful method for studying the composition, thermal properties and dynamics of exoplanet atmospheres, only a few transiting terrestrial exoplanets will be close enough to allow significant transit spectroscopy. Thermal phase curves (variations of the apparent infrared emission of the planet with its orbital phase) have been observed for hot Jupiters in both transiting and non-transiting configurations, and could be observed for hot terrestrial exoplanets. We study the wavelength and phase changes of the thermal emission of a tidally-locked terrestrial planet as atmospheric pressure increases, and address the observability of these multiband phase-curves and the ability to use them to detect atmospheric constituents. We used a 3D climate model (GCM) to simulate the CO2 atmosphere of a terrestrial planet on an 8-day orbit around a M3 dwarf and its apparent infrared emission as a function of its orbital phase. We estimated the signal to photon-noise ratio in narrow bands between 2.5 and 20 microns for a 10 pc target observed with a 6 and 1.5 m telescope (respectively the sizes of JWST and EChO). We find that tmospheric absorption bands produce signatures in what we call the variation spectrum. Planets with no atmosphere produce large variations and can be easily distinguished from dense absorbing atmospheres. Photon-noise limited spectro-photometry of nearby systems could allow us to detect and characterize the atmosphere of non-transiting terrestrial planets known from radial velocity surveys. Two obvious impediments for these types of observations are the required photometric sensitivity (1E-5) over the duration of at least one orbit and the intrinsic stellar variability. However, overcoming these obstacles would give access to one order of magnitude more targets than transit spectroscopy.< Réduire