We use propensity score matching (PSM) to investigate the evolution of the US private sector union wage premium in the mid-1990s. While the literature found a decline in the premium, the use of OLS led to potentially biased estimates. Using PSM to correct for potential bias, we find that although the premium did decline in the mid-1990s, the rate of decline was steeper with more variability in the premium from year to year than found in the literature. Robustness checks indicate that our results are unlikely to be affected by misclassification of union status and selection on unobservables.