We present photometric redshifts for the COSMOS survey derived from a new code, optimized to yield accurate and reliable redshifts and spectral types of galaxies down to faint magnitudes and redshifts out to z ~ 1.2. The technique uses χ2 template fitting, combined with luminosity function priors and with the option to estimate the internal extinction [or E(B - V)]. The median most probable redshift, best-fit spectral type and reddening, absolute magnitude, and stellar mass are derived in addition to the full redshift probability distributions. Using simulations with sampling and noise similar to those in COSMOS, the accuracy and reliability is estimated for the photometric redshifts as a function of the magnitude limits of the sample, S/N ratios, and the number of bands used. We find from the simulations that the ratio of derived 95% confidence interval in the χ2 probability distribution to the estimated photometric redshift (D95) can be used to identify and exclude the catastrophic failures in the photometric redshift estimates. To evaluate the reliability of the photometric redshifts, we compare the derived redshifts with high-reliability spectroscopic redshifts for a sample of 868 normal galaxies with z < 1.2 from zCOSMOS. Considering different scenarios, depending on using prior, no prior, and/or extinction, we compare the photometric and spectroscopic redshifts for this sample. The rms scatter between the estimated photometric redshifts and known spectroscopic redshifts is σ(Δ(z)) = 0.031, where Δ(z) = (zphot - zspec)/(1 + zspec) with a small fraction of outliers (<2.5%) [outliers are defined as objects with Δ(z) > 3σ(Δ(z)), where σ(Δ(z)) is the rms scatter in Δ(z)]. We also find good agreement [σ(Δ(z)) = 0.10] between photometric and spectroscopic redshifts for type II AGNs. We compare results from our photometric redshift procedure with three other independent codes and find them in excellent agreement. We show preliminary results, based on photometric redshifts for the entire COSMOS sample (to i < 25 mag).