Lateral variation in soil N content was studied in five pastures to find the best method of sampling for precise estimates of changes in soil N content with time. Variation increased with sampling area up to, but not beyond, 1.2 ‐ 2.4 ha. Consequently, the most efficient method of sampling for changes in N content was to repeatedly sample a fixed set of small sites. An allocation of 1–3 cores per site was shown to be optimum. The number of cores needed to measure a change in N content in a field with a probability of 0.8 that the 0.95 confidence limits of the mean change would be within ± 50 kg N ha‐1ranged from 150 to 1000.