One of the major challenges for the moment tensor determination is associated with the relatively low‐magnitude events () recorded by few regional stations at relatively large distances (300–600 km) and analyzed with standard velocity models of the region. Difficulties arise from the fact that synthetics in standard models (e.g., those routinely used in the location) cannot properly match real waveforms and favor the appearance of unmodeled time shifts and amplitude discrepancies (e.g., if VMs are constructed to minimize location residuals, they are not sensitive to uppermost shallow layers, which are insufficiently sampled by rays if shallow sources are missing). The situation is even worse when real waveforms can be matched but the retrieved focal mechanism is incorrect. This article investigates an alternative methodology that is more robust with respect to inappropriate velocity models: the inversion of waveform envelopes. The method is built on an empirical basis. It studies the effects of velocity models on synthetic waveforms and finds that the information about focal mechanism is encoded in the variation of the envelope shapes and amplitudes among the seismogram components. Besides synthetic tests, the method has been tested on real data comprising two earthquakes in Brazil: the 2010 4.3 Mara Rosa (MR) and the 2017 4.3 Maranhão earthquakes. When compared with solutions from previous studies, based on many polarities and ad hoc path‐specific velocity models, we obtained in both cases the same mechanism with a single 1D model and a single‐station polarity constraint. The envelope inversion is a promising technique that might be useful in similar sparse networks, such as the one in Brazil, where standard waveform inversion, in general, is not fully efficient.