To improve U-Th zircon dating, we optimized a U-Th-Pb double-dating strategy for young zircon (150–300 ka). We found that the overall U-Th age spectrum is consistent whether assuming a constant melt composition or constant U/Th fractionation between zircon and melt, but testing the representability of the measured glass with the youngest isochron intercept proved essential. A Bayesian model with a uniform prior distribution gave the most accurate estimates of eruption timing for U-Th datasets.