Confirming the precision agriculture hypothesis for variable-rate N applications (VRAs) is challenging. To confront this challenge, researchers have used increasingly sophisticated statistical models to estimate and compare site-specific crop response functions (SSCRFs). While progress has been made, it has been hampered by the lack of a conceptual framework to guide the development of appropriate statistical models. This paper provides such a framework and demonstrates its utility by developing a heteroscedastic, fixed and random effects, geostatistical model to test if VRA can increase N returns. The novelty of the model is the inclusion of site, spatial, treatment, and treatment strip heteroscedasticity and correlation. Applied to data collected in 1995 from two corn (Zea mays L.) N response experiments in south-central Minnesota, results demonstrate the importance of including site, spatial, treatment, and treatment strip effects in the estimation of SSCRFs. Results also indicate a significant potential for VRA to increase N returns and that these potential returns increase as the area of the management unit decreases. At one location, there was greater than a 95% chance that VRA could have increased profitability if the cost of implementing VRA was less than $14.5 ha -1. At the other location, if implementation costs were less than $48.3 ha-1, there was greater than a 95% chance of increased profitability.