Improved understanding of year-to-year late-spring soil nitrate test (LSNT) variability could help make it more attractive to producers. We test the ability of the Root Zone Water Quality Model (RZWQM) to simulate watershed-scale variability due to the LSNT, and we use the optimized model to simulate long-term field N dynamics under related conditions. Autoregressive techniques and the automatic parameter calibration program PEST were used to show that RZWQM simulates significantly lower nitrate concentration in discharge from LSNT treatments compared with areas receiving fall N fertilizer applications within the tile-drained Walnut Creek, Iowa, watershed (>5 mg N L-1 difference for the third year of the treatment, 1999). This result is similar to field-measured data from a paired watershed experiment. A statistical model we developed using RZWQM simulations from 1970 to 2005 shows that early-season precipitation and early-season temperature account for 90% of the interannual variation in LSNT-based fertilizer N rates. Long-term simulations with similar average N application rates for corn (Zea mays L.) (151 kg N ha-1) show annual average N loss in tile flow of 20.4, 22.2, and 27.3 kg N ha -1 for LSNT, single spring, and single fall N applications. These results suggest that (i) RZWQM is a promising tool to accurately estimate the water quality effects of LSNT; (ii) the majority of N loss difference between LSNT and fall applications is because more N remains in the root zone for crop uptake; and (iii) year-to-year LSNT-based N rate differences are mainly due to variation in early-season precipitation and temperature. Copyright ?? 2010 by the American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America. All rights reserved.