Deformation analysis plays an important role for human safety, so investigating the reliability of the obtained results from deformation analysis is extraordinarily important. It uses statistics most widely and if H-0 hypothesis is rejected in applying of the global congruency test, the localisation process is performed to detect one or more displaced points. There are a lot of methods for localisation and they do not have the same and correct results for all cases. Their reliabilities change according to samples, the numbers of displaced and stable points in the network, and the magnitudes of the displacements. There are two reasons for the unsuccessful results of the conventional deformation analysis (CDA) methods: the first is the spreading effect of least squares estimation (LSE); the second is the failure of F-test. LSE is an optimal estimator when observations come from normal distribution, i.e. there is not any outlier in the data set. However, if the observations are non Gaussian, i.e. if there are outliers in the observations or displaced points in the deformation monitoring network, the results obtained from LSE are not optimal. LSE spreads the spoiling effects of deflecting from the assumed model to the residuals of good observations; also it spreads the effects of displaced points on the other estimated points coordinates that are not displaced. Therefore, obtained results diverge from their optimum values. In this study, the results of the CDA have been investigated in the Global Positioning System (GPS) network. To this aim three devices that could move horizontally were built and point displacements were simulated. Eight different scenarios at the regional GPS networks were carried out. The analyses of the GPS measurement were achieved for both whole network and subnetwork designs, so that the effect of the subnetwork design on the deformation analysis was investigated. The analyses of the GPS network show that subnetwork design has a more reliable result than whole network design.