Remotely sensed images in isolation are not usually a subject of interest, but become so if they are related to a property of interest at the ground. To obtain information on the property of interest it is usually necessary to establish a relation (for example, regression) between selected image pixels and a sample of the target property, and apply the relation to the image to estimate the target property at the unobserved places. Where the sample of the target property is obtained by interpolation, the estimates may be smoothed and the regression relation may be biased. In this Letter a potential solution to the problem is presented. The procedure is illustrated with a simple example in which the objective is to model the relation between snow depth in the U.K. and SSM/I microwave imagery.