Deformation monitoring is a very important issue in engineering structures and studying ground deformation. So creating microgeodesy networks in the area of study is required. Detecting unstable points of network is an important issue in microgeodesy networks. Among the classical methods of identifying unstable points of the network, the global congruency test and L1-norm minimization can be referred. The results in the research done in this regard show that when displacement point vector is small, the number of true detection of displaced points using common ways of deformation analysis is less than the number of points that really displaced. The reason can be the spreading nature of the least squares estimation. Now according to the recent research results, in the field of detecting unstable points of network, the idea of using subnetwork analysis is proposed to confront this limitation. In this case the deformation monitoring network is divided into some subnetworks including a subject point and the other source points. In fact according to the unstable points, there will be sub-networks. In this method in whole network first of all using classical methods, the stable and unstable points are investigated. Then by dividing the whole network to subnetworks, every network will be adjusted and its unstable points will be detected. So the relation between unstable points is cutoff and the spreading effect of the least squares is decreased. In this paper this method is evaluated in a simulated and a real network. The results show that using subnetwork analysis compared to global congruency test will improve the results in the correct detection of unstable points. On average this improvement is about 35% in all of the stimulated states. In return improving the results of subnetwork method to L1- norm minimization is about one percent that it cannot be acceptable. In the following the algorithms of detecting unstable points in common methods and the method of analyzing subnetwork were conducted on a real network and the results are consistent with the results of simulated network.