A method and system of crowd-sourced training of a neural network for mobile device indoor navigation and positioning. The method, executed in a processor of a server computing device, comprises: based on RSS parameters acquired at a mobile device from a wireless signal source, localizing the mobile device to a first position within indoor area in accordance with a probabilistic confidence level; if the confidence level exceeds a threshold confidence level, adding the RSS parameters in association with the first position to a fingerprint database of the indoor area; and training a neural network implemented in the processor at least in part based on the RSS parameters as added to the fingerprint database.