Knowledge about active radio transmitters is critical for multiple applications: spectrum regulators can use this information to assign spectrum, licensees can identify spectrum usage patterns and better provision their future needs, and dynamic spectrum access applications can leverage such knowledge to pick operating frequency. Despite the importance of transmitter identification the current work in this space is limited and requires prior knowledge of transmitter signatures to identify active radio transmitters. More naive approaches are limited to detecting power levels and do not identify characteristics of the active transmitter. To address these challenges we propose TxMiner; a system that identifies transmitters from raw spectrum measurements without prior knowledge of transmitter signatures. TxMiner harnesses the observation that wireless signal fading follows a Rayleigh distribution and applies a novel machine learning algorithm to mine transmitters. We evaluate TxMiner on real-world spectrum measurements between 30MHz and 6GHz. The evaluation results show that TxMiner identifies transmitters robustly. We then make use of TxMiner to map the number of active transmitters and their frequency and temporal characteristics over 30MHz-6GHz, we detect rogue transmitters and identify opportunities for dynamic spectrum access.