Moment tensor determinations of small earthquakes are quite challenging. It is because their signal‐to‐noise ratio is satisfactory only at frequencies above the microseismic noise peak (∼0.2 Hz), and waveforms can be modeled only up to ∼1–2 Hz at relatively near stations (epicentral distance of a few kilometers). Therefore, the availability of high‐quality waveforms at near stations is the most critical issue. Dense local networks enable solving specific issues like non‐double‐couple (non‐DC) components (e.g., Vavrycuk, 2011), relation with tectonics (Serpetsidaki et al. , 2013), and/or even space–time clustering of moment tensors (Cesca et al. , 2014). On the contrary, records of weak events at very few local stations hardly allow us to get even DC components of moment tensors. For example, focal mechanism of two microearthquakes, M w 0.2 and 0.4, were obtained by the full‐waveform inversion in the frequency range below 4 Hz and distance below 3.7 km by Benetatos et al. (2013). Fojtikova et al. (2010) studied shallow earthquakes of M w 1.2–3.4 at 0.8–1.6 Hz. A probabilistic nonlinear inversion of local waveforms was developed by Weber (2009) and applied to weak events in Hungary. Various constraints for determining focal mechanisms of small local earthquakes monitored by sparse networks were adopted. For example, Nakamura (2002) suggested a joint use of the P ‐ and S ‐wave polarities. Li et al. (2011) combined the phase and amplitude waveform matching with the first‐motion P polarities and average S / P amplitude ratios. Delouis and Legrand (1999) emphasized inclusion of the near‐field Green’s function terms. Pinar et al. (2003) made tests according to which some single‐station inversions are feasible. Busfar and Toksoz (2013) gave examples of focal mechanisms calculated in good velocity models from less than four stations. Godano et al. (2010, 2011) studied …