Statistical tests that compare means are widely known and used; tests that compare variation are less so. However, evaluating performance of a simulation model over a range of results requires both. In precision agriculture, comparing simulated results to measured results is usually done using linear regression. However, many researchers using this test have mixed spatial and temporal variation, which confounds the analysis and usually inflates the performance measure. Further, relationships exist between the correlation coefficient and the slope that need to be acknowledged to interpret the results. Finally, linear regression cannot address whether the spatial structure that exists in the measured data is simulated or not. We propose comparing semivariograms for this latter evaluation and will illustrate the process and results with an existing dataset of yield variation amenable to semivariogram analysis. We will discuss the estimation of measurement error and how it could be included in the tests of model performance. The suite of comparisons should provide a framework for performance testing of models in precision agriculture.