The color-differential astrometry method consists in measuring the variations of the photocenter position with wavelength. This yields spectral information on the spatial structure of an unresolved source, and could allow, among other applications, the direct detection and spectroscopy of extrasolar planets. We give some estimates of signal-to-noise ratio for CDA using the the NGST mid-infrared instrument (MIRI) spectrograph, assuming for the planetary luminosity either black-body models or recent synthetic spectra. Giant Planets around nearby Sun-like stars are potentially observable over a wide-range of orbital distances, if the precision on the measurement is limited by the fundamental noises. Space-based CDA would then be complementary, in terms of star-planet separation, to ground based interferometry and to coronography. This assumes, though, a proper chromatic stability of the instrument. This note introduces some possible methods for instrumental internal calibration, and presents a possible spatial design for a calibration by spatial modulation.