The U.S. Geological Survey Parkfield Dense Seismograph Array (upsar) recorded successfully strong ground motions during the 2004 Parkfield earthquake (M 6.0) and its aftershock series after waiting for 15 years for an anticipated event like this. The array also recorded the 2003 San Simeon earthquake (M 6.5). Because the array covers a very small area (0.45 km2), these data offer some interesting fresh insights into intrasite variations of seismic ground motions. In this article, we study strong-motion data recorded at the upsar from the San Simeon event, the Parkfield event, and its seven aftershocks. We find that the variations of high- frequency ground motions (e.g., >3 Hz) are very considerable. The largest horizontal peak ground acceleration (pga) (P11, 408 cm/sec2) from the Parkfield event is close to three times of the smallest one (P01, 157 cm/sec2); the largest peak response spectrum is even over three times of the smallest one. The shortest station-to-station distance (between P06 and P07) in the array is only 25 m, but three-component pgas of the two stations differ from a factor of 1.5 for the Parkfield event. The coefficient of variation (Cv = σ/mean) of Fourier acceleration is about 50% at frequencies higher than about 3 Hz. We find that Cv depends strongly on frequency, while it is nearly stable for different earthquakes. The significant variation of the high-frequency ground motions seems to be brought about mostly by the local and neighboring topographic effects, which have a larger effect in horizontal than vertical directions. We also calculate the ratio of vertical to horizontal response spectrum. Our plotting shows that the ratio is not sensitive to earthquake magnitude. We compare the observed motions (pga, 5%-damped pseudoacceleration response spectrum [psa] from 0.02 to 5 sec) with estimations from four commonly used prediction equations (Abrahamson and Silva, 1997; Boore et al., 1997; Campbell and Bozorgnia, 2003; and Sadigh et al., 1997). The comparisons indicate that the significant station-to-station variation reduces largely the accuracy with which a site-specific estimation can be predicted. However, the mean of the observations at the upsar compares reasonably well with these estimations.