We deployed a temporary seismic station at a low-noise site in the Nikko district, Japan, in November 1993, and analyzed velocity waveforms of 17 microearthquakes (0.3≤M≤2.1) with small focal distances (<11 km). We compared the initial parts of the P wave velocity seismograms on the same absolute scale, after correcting for the effects of geometrical spreading and radiation pattern. We find that the P wave radiated from an earthquake of a larger final magnitude has a greater amplitude than that from a smaller earthquake, even at a very early stage when the P wave from the smallest event is still rising with an increasing slope. This observation does not necessarily mean that the initiation of the rupture is dependent on the eventual size because anelastically attenuated waveforms of earthquakes with short time duration could show such an apparent trend. Hence we computed synthetic seismograms to test a group of source scaling models in which source time function grows independently of its eventual size, taking into account anelastic attenuation and instrument response. We find that such scaling models do not easily reproduce simultaneously the size dependence of the initial rise and the pulse duration of the observed waveforms, though poor data coverage above M1.5 prevents us from deriving a firm conclusion. A model with size-dependent growth rate is able to match our observations. ¿ 2000 American Geophysical Union