With the demand of location-based services (LBS) increasing, the higher accurate positioning system is needed. Location Fingerprinting (LF) plays an important role in localization, especially for indoor environments. LF system is based on pattern matching and divided into offline stage and online stage. Before constructing the offline model of LF, the Received Signal Strength (RSS) measurements can be picked or reorganized by some techniques. Besides the channel selection method, the transformation method reduces the online computation complexity and improves the positioning performance. The transformations can reorganize the RSSs from wireless channels and project RSSs into a space where the information is refined. This thesis provides the the theoretical and experimental comparison between two classical transformations, Multiple Discriminant Analysis (MDA) and Principle Component Analysis (PCA). More, we adopted an population-based approach to search the expected transformation to be better than the others. The method Particle Swarm Optimization (PSO) puts the transformations in the space in an evolutionary way to search the optimum information. We conduct the experiments in indoor and outdoor environments, which are the NTUEE building and NTU campus based on WLAN networks and the GSM infrastructure ture respectively. The results show that MDA is better than PCA in addition to theoretical analysis. The results show that the proposed method reduces 20.09∼56.87% and 15.57∼56.54% of the mean error and 67% circular error probable (CEP) for only six channels in indoor environments, respectively, as compared to classical transformation methods. More, the outdoor experimental results show that it also reduces 7.23∼28.80% and 7.61∼29.10% of the mean error and 67% CEP.