Cotton lint yield response to nitrogen levels has been studied extensively based on randomized complete block design experiments. In order to estimate the response curve, the most widely used statistical model is the ordinary least squares (OLS) regression model. Yield errors at specific plots conditioning on nitrogen treatments are canceled out by the model. However, statistically OLS estimates are the most efficient only when the yield errors are completely random. In the experiment practice, the yields errors are often spatially correlated across plots, mainly driven by the unobserved (and uncontrolled) soil characteristics in the field. In the presence of spatially non-random errors, spatial econometric models provide more accurate estimates than OLS. This study applies the Spatial Error model to the estimation of cotton yield response to nitrogen. Our data are from field experiments conducted during three crop years from 2012 through 2014 in three separate locations in Mississippi. Results show that the response coefficients estimated by Spatial Error model are significantly different from those of OLS model. Statistical theory and numerical simulation both prove the spatial model outperforms OLS. This study suggests spatial econometric model is more desirable in analyzing cotton field experiment data compared to OLS.