Global Positioning System (GPS) have been widely used in various applications that require high-precision positioning results, such as geodetic control networks and site velocity estimation. Traditionally, site velocities have been determined using relative GPS positioning due to its higher accuracy than that of point positioning. However, the obtained velocity results are thus relative in nature. In recent years, the precise point positioning (PPP) technique, which uses International GNSS Services (IGS) products - precise satellite ephemerides and clocks - to directly estimate site positions in the International Terrestrial Reference Frame (ITRF), has become a promising tool for absolute site velocity estimation. Nevertheless, the PPP-derived site positions have been reported to be partially biased as a result of the fact that the above IGS products are based on the loosely constrained IGS analysis centre (AC) solutions. Therefore, it is an interesting issue to examine the accuracy of PPP-derived site velocities. In this research, we computed a set of PPP-derived velocity solutions for 33 evenly distributed IGS global tracking stations from 5/11/2006 to 31/12/2010, and compared them with their respective quantities defined in the ITRF. During the ITRF2005 period (since 5/11/2006), the velocity differences in the east, north, and up directions were -0.23±0.75, -0.35±0.44, 0.83±1.69 mm/yr, respectively; and after Helmert transformation, they were -0.01±0.52, -0.02±0.38, 0.00±1.68 mm/yr respectively. Only the vertical component was notably improved by the transformation. This indicates that the quality of IGS products improves with time, so the biases that existed in the PPP solutions are gradually diminished. It is concluded that under the current ITRF2005, one can reliably use the PPP technique to obtain highly accurate absolute horizontal site velocities, on the level of sub-mm/yr.