Monitoring vertical land motion at tide gauges is mandatory if absolute or climate related signals in mean sea level are to be derived from coastal tide gauge records. Different glacial isostatic adjustment models provide very different predictions at the required level of a few tenth of millimetres per year accuracy, while other processes that may affect the vertical stability of tide gauges are still more difficult to predict. An alternative approach is to measure the rates of vertical land motions at the tide gauges. Continuous GPS recordings are presently the most practical and accurate technical solution to address such an issue. The most adequate way to handle the GPS processing seems to be a global scale approach in consistency with the size of the issues: geocentric reference frame realization, long-term stability, global climate change. The laboratories CLDG (Centre Littoral De Géophysique) and LAREG (Laboratoire de Recherches en Géodésie) contribute to the Tide Gauge Benchmark Monitoring (TIGA) IGS pilot project since October 2002 by routinely processing a global set of about 140 GPS sites within the ULR analysis centre consortium. The set comprises about 122 TG-CGPS stations that are analysed using a free network approach. Time series of more than six years are now available. The accuracy of the vertical component is very sensitive to the reference frame definition and realization. We therefore performed some tests in order to quantify the impact of various analysis options on the stability of our realization. We compared the stability of our global solutions with respect to ITRF2000 solution and to its IGS realization called IGS00, while changing the number of reference stations and their geometry. The results indicate that the more reference stations are used the better the alignment to ITRF is performed. A large global distribution of the reference frame stations seems to mitigate individual station problems. The reference frame implementation is achieved within several millimetres accuracy on a weekly basis. The tests indicate that the choice of IGS00 or ITRF2000 datum was not a relevant issue to perform the alignment to ITRF at this level of accuracy.