AbstractThe spatial variability of Rhizobium japonicum populations was evaluated for two soils on the Coastal Plain of North Carolina. The Nahunta soil (fine‐silty, siliceous, thermic Arenic Paleudults) was somewhat poorly drained and had an organic matter content of 20 g kg−1 whereas the Pocalla soil (loamy, siliceous, thermic, Arenic Paleudults) was well drained and had 10 g kg−1 organic matter. One hundred and five soil samples from the Ap horizon were collected from an apparently uniform 18‐m by 18‐m area at each site and were subjected to analysis by the plant infectivity—most‐probable‐number (MPN) technique to determine the number of rhizobia. The log10 MPN (log MPN) of R. japonicum varied from 2.63 to 7.03 g−1 soil; mean log MPN values were 4.10 and 6.07 g−1 soil for Pocalla and Nahunta, respectively. Semivariograms for sample separation distances (h) < 2 m at each site were developed from soil samples collected on transects (fences). Semivariograms for h ≥ 3 m were developed using data collected on a 3‐m grid. Analysis of the semivariograms constructed from the log MPN data indicated a different variance structure at each site. Moreover, the variance structure for log MPN was found to be directional, i.e., it was different in the direction “parallel to the row” compared to the direction “perpendicular to the row.” For the Pocalla soil the semivariance in the “down‐the‐row” direction was greater for values of h < 1.05 m than it was for h ≥ 3 m. This apparent discrepancy is thought to be due to selection of the transect from which the 0.2‐m spaced samples were collected in a region of the field which is not typical of the field as a whole. This result demonstrates the importance of using a sampling regime which is not biased toward one particular area of a field. Root densities, inherent soil properties, seasonality, and management practices are possible contributors to the subtle changes in rhizobia populations that were observed.