We investigate the properties of the most optically faint sources in the GOODS-N area (RAB 〉 26.5). These extremely optically faint populations present an uncharted territory although they represent an appreciable fraction of the X-ray sources in the GOODS-N field. The optically faint sources are believed to contain either red active galactic nuclei (AGN) at moderate redshifts or possibly quasi stellar objects (QSOs) at very high redshift. We compile our sample by first finding the 3.6 μm IRAC counterparts of the X-ray sources and in turn by searching for the optical counterparts of the IRAC sources. No counterparts were found for 35 objects in the R-band Subaru optical images. Of these, 18 have HST ACS counterparts, while the remaining have no optical counterparts. The vast majority of our 35 sources are classified as extremely red objects (EROs) on the basis of their V606-KS lower limits. Their derived photometric redshifts show that these populate moderate redshifts (median z ~ 2.8), being at markedly different redshifts from the already spectroscopically identified population which peaks at z ~ 0.7. The Spitzer IRAC mid-IR colours of the sources without HST counterparts tend to lie within the mid-IR colour diagram AGN “wedge”, suggesting either QSO, ultra luminous infrared galaxy (ULIRG; Mrk231) templates or early-type galaxy templates at z 〉 3. A large fraction of our sources (17/35), regardless of whether they have HST counterparts, can be classified as mid-IR bright/optically faint sources (dust obscured galaxies) a class of sources which is believed to include many heavily absorbed AGN. The co-added X-ray spectrum of the optically faint sources is very flat, with a spectral index of Γ ≈ 0.87, significantly flatter than the spectrum of the X-ray background. The optically faint (R 〉 26.5) X-ray sources constitute more than 50 per cent of the total X-ray population at redshifts z 〉 2, bearing important implications for the luminosity function and its evolution. Considering X-ray sources with 2 〈 z 〈 4 we find good agreement with a modified pure luminosity evolution (PLE) model.