The NIR Ca II triplet has proven to be an important tool for quantitative spectroscopy. Here we present results of synthetic spectral analysis for the Ca II triplet for low-metallicity red giant stars, combined with observational data. Our results start to deviate strongly from the widely-used and linear empirical calibrations below [Fe/H] = −2. We provide a new calibration for Ca II triplet studies which is valid down until [Fe/H] = −4 and apply this new calibration to current data sets. We suggest that the classical dwarf galaxies are not so devoid of extremely low-metallicity stars as was previously thought and discuss preliminary results and possibilities for follow-up observations of these extremely low-metallicity candidates.