Proper sampling of grain losses during harvesting operations, with reliable and efficient sizing of sample, is necessary for an efficient adjustment of the harvester to avoid harvest losses. Thus, the objective of this study was to estimate sampling intensity and sample size for harvest loss evaluations in soybean crops. Sampling was carried out in five locations with soybean crops, evaluating three different harvesters. Harvest losses were measured using square wooden frames (50 × 50 cm), which were arranged on the ground longitudinally across the harvester header width after its passage; this process was repeated 25 times at each location. The greatest harvester header width in Location 1 enabled to simulate different sample sizes for this location (50 × 100, 50 × 150, 50 × 200, and 50 × 250 cm). Only one sample size (50 × 100 cm) was used for the simulations in the other locations. Sampling only half the harvester header width is recommended to estimate harvest losses when using 62-foot harvester header, whereas 12.5- and 17-foot harvester headers requires sampling across the entire harvester header width, with a semi-amplitude of the confidence interval of 20% of the mean for all harvesters.