In this paper indoor localization system based on the RF power measurements of the Received Signal Strength (RSS) in WLAN environment is presented. Today, the most viable solution for localization is the RSS fingerprinting based approach, where in order to establish a relationship between RSS values and location, different machine learning approaches are used. The advantage of this approach based on WLAN technology is that it does not need new infrastructure (it reuses already and widely deployed equipment), and the RSS measurement is part of the normal operating mode of wireless equipment. We derive the Cramer-Rao Lower Bound (CRLB) of localization accuracy for RSS measurements. In analysis of the bound we give insight in localization performance and deployment issues of a localization system, which could help designing an efficient localization system. To compare different machine learning approaches we developed a localization system based on an artificial neural network, k-nearest neighbors, probabilistic method based on the Gaussian kernel and the histogram method. We tested the developed system in real world WLAN indoor environment, where realistic RSS measurements were collected. Experimental comparison of the results has been investigated and average location estimation error of around 2 meters was obtained.