The accuracy of GPS (Global Positioning System) derived relative positions of stations depends on several factors. Besides the baseline length and duration of observation session, the methodology and the software used influence the results. In this paper, the observations made in the Marmara Continuous GPS Network (MAGNET) have been analysed to determine how the accuracy of derived relative Positions of GPS stations depends on the baseline length and the duration of the observing session. Seven days of GPS observations in the MAGNET collected in 2002 were processed in the ITRF 2000 reference frame using Bernese 4.2 software. The baseline length varies between 6 km and 340 km, the session duration varies between 4h and 24h. The independent baseline components have been analysed.