Future surveys studying galaxy evolution are going to observe a vast number of galaxies (LSST is going to target billions of galaxies). The redshift of these objects is one the most important quantities we need to estimate in order to better understand our universe. Though spectroscopic redshifts (spec-z's) are more accurate, they are also much harder to obtain, therefore photometric redshifts (photo-z's) are the only option for such a large number of objects. Different codes that estimate photo-z's also produce probability distribution functions (PDFs), from which we can obtain uncertainties associated with point values. While point values, produced by the various participants using these codes, are usually fairly accurate, their uncertainties do not perform well when tested. We use statistical methods to improve the photo-z PDFs, which yield better results both for the point estimates, and the associated errors.