Narrowband Internet of Things (NB-IoT) has quickly become a leading technology in the deployment of IoT systems and services, thanks to its appealing features in terms of coverage and energy efficiency, as well as compatibility with existing mobile networks. Increasingly, IoT services and applications require location information to be paired with data collected by devices; NB-IoT still lacks, however, reliable positioning methods. Time-based techniques inherited from Long Term Evolution (LTE) are not yet widely available in existing networks, and are expected to perform poorly on NB-IoT signals due to their narrow bandwidth. This investigation proposes a set of strategies for NB-IoT positioning, based on fingerprinting, that use coverage and radio information from multiple cells. The proposed strategies are evaluated on a large-scale dataset that includes experimental data from two NB-IoT operators. Results show that the proposed strategies, using a combination of coverage and radio information from multiple cells, outperform current state-of-the-art approaches based on single cell finger-printing, with a minimum average positioning error of about 20 meters, consistent across different network scenarios, vs. about 70 meters for current state-of-the-art.