Assessing and modelling within-field variability of crop and soil at different scales is a pre-requisite for effective management in Precision Agriculture. Actually, site-specific agronomic management cannot disregard knowledge of the associated scale(s) of the phenomena occurring in the soil and plant that a farmer wishes to control. A partition of the field into management zones has then to be scale-dependent because it may vary as a result of scaling. The objective of this study was to construct a scale-dependent model of spatial soil-crop relationships by using a multivariate geostatistical approach for producing field partition at the relevant scales. Some soil attributes and confined compression function parameters, photosynthetic parameters and wheat yields were determined at 100 locations on a regular grid (150 m×150 m) covering the whole field area (200 ha). A nested linear model of coregionalization was estimated and a Factorial cokriging analysis was performed to model the multivariate correlation structure and calculate regionalized factors at three different scales. The retained regionalized factors were: the first factor at 545m scale, showing greater spatial correlation with yield parameters; the second factor at the same scale, more correlated with plant photosynthetic parameter and mechanical properties of soil; the first longer-scale factor, which might be interpreted as an inverse indicator of soil compaction . It was more related to variables clay and bulk density that can be assumed stationary at a scale extending beyond the actual size of the field. Each of these three factors produced a different partition of the field, each of which could be used for different purposes and to manage distinct agronomic operations. The results showed the complexity of the soil-crop interactions due to the influence of distinct sources of variation working on several spatial scales. The multi-scale delineation of the field in homogeneous zones emphasises the nee
