We here report a method for synthesizing design earthquake motions at a short epicentral distance that makes use of microearthquake records. Characteristics such as the corner frequency and flat level were analyzed from displacement spectra determined from the digitized time histories of microearthquakes. A statistical model of microearthquake motions then was constructed by applying the nonstationary second-order autoregressive moving average (AR-MA) process. The attenuation laws of the AR-MA parameters were derived from a regression analysis. A waveform that satisfied these attenuation laws was regarded as a statistical Green function. Strong earthquake motions were synthesized by superposing the statistical Green function. In order to make this superposition, a new scaling law was derived taking into account the difference of stress drop between large events and microearthquakes. Peak accelerations and response spectra of synthesized accelerograms in the region near the source were large in comparison to the values which had ever proposed.