We present a method for calibrating seismic networks in order to retrieve highly accurate seismic moment tensors. The method is based on a joint inversion of large datasets of earthquakes for moment tensors and for amplifications of the network stations, which encompass the instrumental amplifications of sensors and the gain factor of the acquisition system, as well as the local site effects neglected in modeling of Green’s functions. The method is capable of detecting the reverse polarity of sensors, incorrect orientation of sensors, or anomalous site effects caused by local geological conditions at individual stations. The robustness and accuracy of the method are tested on synthetic data with different noise levels, station configurations, and a variety of focal mechanisms. The numerical modeling confirms that the inversion code works well and yields robust results. The tests show that the moment tensors, calculated from data of properly calibrated seismic networks, are significantly more accurate. Finally, the method is applied to observations in West Bohemia, Czech Republic, in order to calibrate a network of 22 local seismic stations operated in this region and to calculate accurately the double‐couple (DC) and non‐double‐couple (non‐DC) components of moment tensors of 200 selected micro‐earthquakes. The results indicate that the method is efficient and can easily be used to calibrate other networks. For example, it can be used in inverting laboratory data, where the coupling effects between the sensor and a rock sample are difficult to quantify, or mining and borehole data, where the calibration and orientation of the sensors are frequently unknown. Moreover, the method can be applied to all studies, which deal with retrieving and interpreting highly accurate moment tensors and their DC and non‐DC components.