Abstract We analyze high sampling waveforms of the initial part of P-wave recorded at the 1800-m-deep borehole seismographs at the Nojima fault from December 1999 to May 2000 to clarify the initial rupture process of microearthquakes. We select 12 events with high S/N, whose magnitudes range from −0.3 to 2.2 and hypocentral distances from 1 to 11 km. We adopt the two different source models by Sato and Hirasawa (1973) and by Sato and Kanamori (1999). The former (model by Sato and Hirasawa (SH model)) generates only a ramp-like onset of velocity pulse. The later (model by Sato and Kanamori (SK model)) is able to generate a weak initial phase that is controlled by a trigger factor and the length of pre-existing crack. We perform the waveform inversion to estimate the optimum source parameters of each model. Waveforms of 5 of the 12 events are clearly reproduced by both SH model and SK model with a large trigger factor and a small length of pre-existing crack. The others are explained by not SH model but only SK model with a small trigger factor and a large length of the pre-existing crack, indicating that the weak initial phase is a nucleation phase and reflects the source process. These seven events satisfy roughly a relation that a large event has a large length of the pre-existing crack; the final crack length is proportional to the length of the pre-existing crack.