According to certain aspects, the invention provides a method of achieving near-fingerprinting accuracy without fingerprinting database costs, via parametric representation of training data. In embodiments, this is achieved by using a parametric representation for the training data and limiting the region of applicability of the parametric model in order to mimic fingerprinting. According to certain additional aspects, an algorithm according to principles of the invention includes two separate parts, training and positioning. Training encompasses the derivation of the parametric model and region of applicability. This comprises access point position, propagation model and the window of applicability. Positioning covers the use of the derived model along with a set of fresh observations to generate a position estimate.