Using Japanese economic data and a Monte Carlo simulation, this study analyzes the consequences of ignoring deterministic trends in mixed unit-root data for Granger noncausality tests. Results from an augmented VAR suggest over-rejection in certain empirically relevant cases at various sample sizes.