A priori knowledge of spatial crop yield distribution in farmers' fields is important for adequate application of fertilizers and for avoiding nutrient losses, e.g., as nitrous oxide emission or nitrate leaching due to over-fertilization. In years with different weather conditions during early crop growth stages, local site conditions cause annually varying patterns of biomass development resulting in differing spatial crop yield variability within the same field. The availability of crop sensors provides an opportunity to consider the spatially varying biomass development at early stages during a growing season for yield predictions. Besides identifying suitable growing stages at which the crop stand should be monitored, a remaining question is the scale of observations and data aggregation for adequate yield prediction. The objective of this study was to quantify local variability and the spatial variability structure at different data aggregation levels of two crop sensors and of winter wheat grain yield. In a field, vehicle-borne measurements of the normalized difference vegetation index (NDVI), red edge inflection point (REIP), and a mechanical pendulum sensor were taken in spring of 2003 in a wheat field in Luettewitz, Germany. Sensor and crop yield observations were obtained for approximately every 5 m in the driving direction during monitoring and harvest. Raw data were processed and aggregated over circular areas with 10, 15, and 20 m radius. The overall variance decreased with increasing area size. While for crop yield, the nugget variance hardly changed with area size, it was reduced with decreasing aggregation area for NDVI, REIP and pendulum measurements. Even for relatively small areas (radius of 10 m) a strong spatial association between sensor measurements taken in spring and grain yield was observed. This result of well defined spatial covariance was successfully used in state-space models for crop yield prediction.