Abstract A space telescope capable of high-contrast imaging has been recognized as the avenue toward finding terrestrial planets around nearby Sun-like stars and characterizing their potential habitability. It is thus essential to quantify the capability of reflected light spectroscopy obtained through direct imaging for terrestrial exoplanets, and existing work focused on planetary analogs of modern Earth. Here we go beyond Earth analogs and use a Bayesian retrieval algorithm, ExoReL R , to determine what we could learn about terrestrial exoplanets from their reflected light spectra. Recognizing the potential diversity of terrestrial exoplanets, our focus is to distinguish atmospheric scenarios without any a priori knowledge of the dominant gas. We find that, while a moderate-resolution spectrum in the optical band (0.4−1.0 μm) may sufficiently characterize a modern Earth analog, it would likely result in incorrect interpretation for planets similar to Archean Earth or having CO2-dominated atmospheres. Including observations in the near-infrared bands (1.0−1.8 μm) can prevent this error, determine the main component (N2, O2, or CO2), and quantify trace gases (H2O, O3, and CH4) of the atmosphere. These results are useful to define the science requirements and design the wavelength bandwidth and observation plans of exoplanet direct imaging missions in the future.