The recently developed MM5 three-dimensional variational (3DVAR) assimilation system at NCAR has been used to investigate influences of ingested sounding and ship observation data on numerical simulations of severe weather in the vicinity of Taiwan. Three weather events were simulated in this study, which include a Mei-Yu front in June 1998, Supertyphoon Bilis in August 2000 and Typhoon Nari in September 2001. For the Mei-Yu front, the simulated low-pressure system northeast of Taiwan is stronger when the 3DVAR is performed during initialization. The simulated patterns of heavy rainfall just off the southern tip of Taiwan are also closer to observations for the run with 3DVAR. For the second case of Typhoon Bilis, both runs with and without 3DVAR show a northward bias in track upstream of eastern Taiwan. However, due to less northward track deflection, the simulated heavy rainfall in Taiwan for the 3DVAR run is in better agreement with observations. For the no-3DVAR run, the simulated Typhoon Nari consistently moves southwestward toward Taiwan but then makes an incorrect landfall at northwestern Taiwan. With 3DVAR, the track simulation is improved with a landfall position at northeastern Taiwan. During landfall, the associated cloud convection is enhanced as the intense vortex core is in confrontation with the leading edge of the Central Mountain Range (CMR) and its movement is slowed down along the northwestern coast. The combination of both compression and stagnation of the embedded convective system may explain the extremely intense rainfall in northern Taiwan. The feature of observed intense rainfall over the southwestern slope base of the CMR is also captured, but the intensity is considerably underpredicted due to lagging and weakening of the vortex core at later times for both runs with and without 3DVAR. The track simulation at later times has also been improved in the 3DVAR run, but the associated geometric distributions of 72-h accumulated rainfall amounts in general are similar to those without 3DVAR. In the typhoon experiment with a bogus vortex, the 3DVAR run preserves the initial vortex intensity quite well. However, the rainfall and track simulations are not improved by the ingestion of the bogus vortex for both runs with and without 3DVAR.