We mapped and analyzed two vertical exposures—exposed on the walls of a 3- to 5-m-deep, 70-m-long excavation and a smaller 3-m-deep, 10-m-long excavation—across the 1999 rupture of the Chelungpu fault. The primary exposure revealed a broad anticlinal fold with a 2.5-m-high west-facing principal thrust scarp contained in fluvial cobbly gravel beds and overlying fine-grained overbank deposits. Sequential restoration of the principal rupture requires initial failure on the basal, east-dipping thrust plane, followed by wedge thrusting and pop-up of an overlying symmetrical anticline between two opposing secondary thrust faults. Net vertical offset is about 2.2 m across the principal fault zone. From line-length changes, we estimate about 3.3 m of horizontal shortening normal to fault strike. The ratio of these values yields a total slip of 4.0 m and an estimate of about 34° for the dip of the fault plane below the excavation. This value is nearly the same as the 35° average dip of the fault plane from the surface to the hypocenter. Restoration of the exposed gravelly strata and adjacent overbank sediments deposited prior to the 1999 event around the principal rupture suggests the possible existence of a prior event. A buried 30-m-wide anticlinal warp beneath the uplifted crest of the 1999 event is associated with three buried reverse faults that we interpret as evidence for an earlier episode of folding and faulting in the site. The prior event is also recorded in the smaller excavation, which is located 40 m south and is oriented parallel to the larger excavation. Radiocarbon dating of samples within the exposed section did not place tight constraints on the date of the previous event. Available data are interpreted as indicating that the previous event occurred before the deposition of the less than 200 14C yr B.P. overbank sands and after the deposition of the much older fluvial gravels. We interpret the previous event as the penultimate event relative to the 1999 Chi-Chi earthquake. We estimated the long-term slip rate of the Chelungpu fault to be 10-15 mm/yr during the last 1 Ma, based on previously published retrodeformable cross sections. This rate is, however, significantly higher than geodetic rates of shortening across the Chelungpu thrust where two pairs of permanent Global Positioning System stations suggest 7-10 mm/yr of shortening across the fault. Given the 4 m of average slip, the long-term slip rate yields an interseismic interval of between 267 and 400 yr for the Chelungpu fault.