A waveform inversion scheme was developed in order to explore the resolving power of one or two seismic recordings at short epicentral distance for the determination of focal mechanisms and the identification of the fault plane of earthquakes. Two key features are used to constrain the fault parameters with a reduced number of stations: (1) a simple finite-dimension source model and (2) the modeling of the complete displacement field, including the near-field waves. The identification of the fault plane should be possible, even with a single station, as soon as the seismograms produced by the two nodal planes of a same focal mechanism are significantly different, which is the general case when waveforms are controlled by source finiteness. Seven parameters, including the strike, dip, rake, and dislocation, are explored with a grid search, and the minima of the misfit error between the observed and calculated seismograms are mapped. With such an approach, it is possible to conclude about the uniqueness or nonuniqueness of the solutions. The method is tested with three earthquakes of moderate to large size for which the fault plane is well established and for which strong-motion records are available at maximum distances of a few tens of kilometers. Test events are the 1994 Northridge (Mw = 6.7, California), the 1996 Copala (Mw = 7.3, Mexico), and the 1996 Pinotepa Nacional (Mw = 5.4, Mexico) earthquakes. In the case of inversions with two stations, we find a unique solution, or a group of similar solutions, with a good estimation of the focal mechanism and the proper selection of the fault plane. Our results also show that in some cases a single station may be enough to recover the fault parameters. The inversion scheme presented here may be systematically applied to future earthquakes, especially to those recorded by few stations. It should be particularly useful in the case of blind faults for which the fault plane may not be identified with the help of other data.