The current paper tests the export-led growth hypothesis using Taiwan's time series data from 1951 to 1998 within the framework of four-variable vector autoregressive (VAR) models that contain unit roots. Two testing procedures are employed. One is proposed by Toda and Phillips (1993, 1994), which employs a Johansen-type error correction model. The other is a lag-augmented VAR approach, proposed by Toda and Yamamoto (1995), which need not take into account possible integration and cointegration among series. Empirical results from both approaches appear to reject the null hypothesis of Granger no-causality from export to output. However, no causal link from output to export is found.