Crop yields are largely determined by climate variations and agricultural management practices, such as irrigation, fertilization and residue management. Understanding the role of these factors in regulating crop yield variations is not only important for improved crop yield production, but also equally valuable for future crop yield prediction and food security assessments. Recently, the Community Land Model (CLM) has been augmented and evaluated for simulating corn, soybean and cereals at coarse aerial resolutions of 2 degrees (2000x2000 km). To better understand the underlying mechanisms controlling yield variations, we implemented and validated the agricultural version of CLM (CLM-crop) at a 0.125 degree resolution over the Conterminous United States (CONUS). We conducted a suite of numerical experiments to untangle the relative influence of climatic variations (temperature, precipitation, and radiation) and agricultural management practices on yield variations for the past 30 years at the US county level. Preliminary results show that the model with default parameter settings captures well the temporal variations in crop yields, as compared with the actual yield reported by the US Department of Agriculture (USDA). However, the magnitude of simulated crop yields is substantially higher, especially in the Mid-western US. We find that improved characterization of fertilizers and irrigation practices is key to model performance. Retrospectively (1979-2012), crop yields are more sensitive to changes in climate factors (such as temperature) than to changes in crop management practices. The results of this study advances understanding of the dominant factors in regulating the crop yield variations at the county level, which is essential for credible prediction of crop yields in a changing climate, under different agricultural management practices.