Real-time canopy sensor technology can provide detailed information about crop development and/or stress. Measurement detail (measurements per unit area) can be varied by sensor physical density (number of sensors along the applicator boom) and by sensor data acquisition rates. Increased detail by increased sensor physical density also increases the cost of the equipment. Efficient use of real-time canopy sensors to effect variable rate application of nitrogen (N) requires knowledge of the scale (resolution) of the variation in crop stress as measured by the sensor. The ideal resolution has been called the “optimum field element size”, defined as the area providing the most precise measure of a parameter with a spatially variable magnitude. This “element” has been calculated taking into account the area at which a “cause and effect” relationship could be measured. However, these relationships, and the associated optimal areas, may not be identical among fields and across production seasons. In this study, the “optimum field element size” was estimated using the spatial structure of the measured characteristic that was being used to calculate N input requirements. Knowing the amount of optical data needed for efficient N fertilizer application requires an estimate of the most efficient combination of physical sensor density (number of sensors along the applicator boom) and sensor output density (sensor readings per unit distance/area). Reducing the density of sensors and their output will reduce the capital cost of such N applicators. The objective of this study was to test sensor number and sensor sampling rate (sample grid size) to adequately describe field variation in wheat (Triticum aestivum L.) canopy NDVI (normalized difference vegetative index) values. NDVI data were collected on wheat at the Feekes 3 (Zadoks Stage 26) growth stage in two fields in central and western Kentucky, in both 2004 and 2005, with a GreenSeeker ^TM (N Tech Industries Inc, Ukiah, CA) variable N rate sensor/applicator (U.S. Patent No. 5389781). Spatial structure of each site-year's NDVI data was described by a variogram. Experimental variograms (local NDVI variograms) were computed from a neighbourhood of 100 observations within a mean radius that depended on field size. The exponential model was fitted to all semivariograms to facilitate comparison of semivariogram parameters. Tested grid sizes ranged from 0.56 to 5.06 m². Variograms for high density data sets were compared with those obtained with fewer sampling points (greater grid size). In evaluating the consistency in spatial structure for a given site-year, differences in variogram parameters due to grid size/sampling density that were 10% or less were considered negligible. NDVI values from each individual sensor were analyzed. It was assumed that characteristics of the NDVI data population measured by each sensor were similar for each site-year because the sensors were in close proximity. Average NVDI readings and NDVI spatial structure (variogram parameters) varied with site-year. However, in all cases it was possible to increase grid size from 0.56 m² to 5.06 m² without significantly affecting a field's NDVI spatial structure. Nugget, range and sill values were maintained across evaluated grid sizes. We conclude that it is possible to increase the effective grid size to 5.06 m², both by decreasing the number of sensors along the toolbar and by increasing the sensor measurement time interval. Wheat growers using this technology could achieve optimal characterization of variation in NDVI with significant reduction in fixed machinery cost (fewer sensors). This enables wheat growers to apply N fertilizer with larger, less expensive (per unit of application width) machines. Given observed differences in sensor performance, sensor uniformity remains an issue. The effect(s) of differences among sensors might, if large enough, override the effect(s) of increasing or decreasing NDVI sampling density.