Wireless Local Area Network (WLAN) fingerprinting methods based on 802.11 signal strength are becoming increasingly the dominating indoor positioning techniques, due to their independence from radio propagation models and cost-effectiveness in terms of hardware and deployment requirements. However, frequent environmental changes cause inconsistency between the fingerprints stored in the radio map and the current radio behavior, thus jeopardizing their accuracy. Re-calibration of the area for updating the radio map incurs considerable amount of time and manual effort. In this paper, we aim to overcome this limitation by adapting to the new radio characteristics through user cooperation and thus eliminating the need of re-calibration. To that end, we propose a cooperative learning algorithm, whereby users exchange their real-time signal measurements in order to refine their estimated locations. The refinement process relies on the neural network structure of self-organizing map (SOM) which is of special interest for localization due to the key property of its neurons in self-organizing in geographic structures based on their similarity to a high-dimensional input data. In our solution, each user is regarded as a neuron of its local SOM network and runs in distributed fashion a modified version of SOM learning algorithm by considering its signal relationship with its neighboring users. Performance evaluation results demonstrate accuracy improvement over both the baseline deterministic and probabilistic fingerprinting approaches, while keeping the communication and computational overheads low