This paper describes the upgrade of the GOCE Level 1b gradiometer processing as part of ESA's Payload Data Segment (PDS). Four processing steps have been identified which can be improved: 1. The optimal determination of the angular rates of the satellite, based on a combination of star sensor and gradiometer data. This is the so-called angular rate reconstruction. 2. The optimal determination of the spacecraft's attitude, again based on a combination of star sensor and gradiometer data. 3. The combination of data of all simultaneously available star sensors. And, 4. the calibration of the measured accelerations is improved by taking the time dependence of selected calibration parameters into account. In this paper the complete upgrade of the gradiometer Level 1b processing is described, and the expected improvement of the gravity gradients as well as the gravity field solutions is quantified. The largest overall improvements are due to the new method for angular rate reconstruction, mainly at the low to medium frequencies and at the harmonics of the orbital revolution frequency. In addition, spurious artifacts in the gravity gradient V _yy , which are caused by non-perfect common mode rejection, can be reduced significantly by the improved calibration approach. We show that the standard deviation of the gravity gradient tensor trace can be reduced by about 90 percent for the frequencies below the gradiometer measurement band (i.e. below 5 mHz) and by about 4 percent within the measurement band (from 5 to 100 mHz). The geoid error of satellite gravity gradiometry solutions based on 61 days of data is reduced from 3.0 to 2.2 cm between spherical harmonic degree 20 and 150. The corresponding gravity anomaly error, between the same degrees, is reduced from 0.7 to 0.5 mGal.