Reliable agronomic and fibre quality data generated in Upland cotton (Gossypium hirsutum L.) cultivar performance trials are highly valuable. The most common strategy used to generate reliable performance trial data uses experimental design to minimize experimental error resulting from spatial variability. However, an alternative strategy uses a posteriori statistical procedures to account for spatial variability. In this study, the efficiency of the randomized complete block (RCB) design and nearest neighbour adjustment (NNA) were compared in a series of cotton performance trials conducted in the southeastern USA to identify the efficiency of each in minimizing experimental error for yield, yield components and fibre quality. In comparison to the RCB, relative efficiency of the NNA procedure varied amongst traits and trials. Results show that experimental analyses, depending on the trait and selection intensity employed, can affect cultivar or experimental line selections. Based on this study, we recommend researchers conducting cotton performance trials on variable soils consider using NNA or other spatial methods to improve trial precision.