The 5 February 2016, Meinong, Taiwan, earthquake brought extensive damage to nearby cities with significant pulse‐like velocity ground motions. In addition to the spatial slip distribution determination using filtered strong‐motion data, we show that, with the advantage of the densely distributed seismic network as a seismic array, we can project the earthquake sources (asperities) directly using nearly unfiltered data, which is crucial to the understanding of the generation of the pulse‐like velocity ground motions. We recognize that the moderate but damaging 6.6 Meinong earthquake was a composite of an 5.5 foreshock and an 6.18 mainshock with a 1.8–5.0 s time delay. The foreshock occurred at the hypocenter reported by the official agency, followed by the mainshock with a centroid located at 12.3 km to the north‐northwest of the hypocenter and at a depth of 15 km. This foreshock–mainshock composition is not distinguishable in the finite‐fault inversion because it filtered the seismic data to low frequencies. Our results show that the pulse‐like velocity ground motions are mainly attributed to the source of mainshock with its directivity and site effects, resulting in the disastrous damages in the city of Tainan. Although finite‐fault inversion using filtered seismic data for spatial slip distribution on the fault has been a classic procedure in understanding earthquake rupture processes, using a dense seismic network as a seismic array for unfiltered records helps us delineate the earthquake sources directly and provide more delicate information for future understanding of earthquake source complexity.