The effects of station distribution and focal mechanism on moment tensor inversion are examined by means of numerical experiment. The far-field displacements of P and Rayleigh waves computed from fault models are inverted to the moment tensor after the error corresponding to the observation error is added to the displacement. The rms error is defined as rms of the differences between the given fault parameters and the parameters obtained from the inverted moment tensor. Experiments are made for both ideal and actual station coverages. The rms errors are found to be proportional to the random error added. For the ideal station coverages, the rms errors are smaller for strike-slip type events than for dip-slip types when only P waves are used. Both types give similar errors when only Rayleigh waves are used. For the actual station distribution of the SRO/ASRO network and earthquakes all over the world, the rms errors vary significantly with the location and mechanism of the events. The minor double couple, which must vanish if no errors are added, reaches 30 per cent of the major double couple for some events when the standard deviation of the random error is 10 per cent of the amplitude. These experiments indicate that the station coverage causes serious effects on moment tensor inversion, and we must be careful to interpret the routinely determined results. Furthermore, uniform station distribution of the global seismograph network is desirable.