It is generally agreed upon that size invariance and the absence of spurious resolution are two requirements that characterize well behaved spatial samping in visual systems. We show that these properties taken together constrain the structure of receptive fields to a very large degree. Only those field structures that arise as solutions of a certain linear partial differential equation of the second order prove to satisfy these general constraints. The equation admits of complete, orthonormal families of solutions. These families have to be regarded as the possible receptive field families subject to certain symmetry conditions. They can be transformed into each other via unitary transformations. Thus a single representation suffices to construct all the others. This theory permits us to classify the possible linear receptive field structures exhaustively, and to define their internal and external interrelations. This induces a principled taxonomy of linear receptive fields.