A large population of heavily obscured, Compton Thick AGNs is predicted by models of galaxy formation, models of Cosmic X-ray Background and by the ``relic'' super-massive black-hole mass function measured from local bulges. However, so far only a handful of Compton thick AGNs have been possibly detected using even the deepest Chandra and XMM surveys. Compton-thick AGNs can be recovered thanks to the reprocessing of the AGN UV emission in the infrared by selecting sources with AGN luminosity's in the mid-infrared and faint near-infrared and optical emission. To this purpose, we make use of deep HST, VLT, Spitzer and Chandra data on the Chandra Deep Field South to constrain the number of Compton thick AGN in this field. We show that sources with high 24$μ$m to optical flux ratios and red colors form a distinct source population, and that their infrared luminosity is dominated by AGN emission. Analysis of the X-ray properties of these extreme sources shows that most of them (80$±15%$) are indeed likely to be highly obscured, Compton thick AGNs. The number of infrared selected, Compton thick AGNs with 5.8$μ$m luminosity higher than $10^{44.2}$ erg s$^{-1}$ turns out to be similar to that of X-ray selected, unobscured and moderately obscured AGNs with 2-10 keV luminosity higher than $10^{43}$ erg s$^{-1}$ in the redshift bin 1.2-2.6. This ``factor of 2'' source population is exactly what it is needed to solve the discrepancies between model predictions and X-ray AGN selection. ; Comment: Revised version, to be published by The Astrophysical Journal