The development of digital retinal color imaging causes a substantial increase in the number and the size of retinal image databases. Image processing methods have been developed to help the specialists analyze these images. However, the heterogeneity of the databases, regarding image scale, contrast or quality, makes the design of generic image processing algorithms difficult. The presented work focuses on the spatial normalization of these images. The method is based on the definition of a size invariant in images. Unfortunately, the size of anatomical structures is either difficult to measure (e.g. the distance between optic disk and fovea requires a tricky segmentation of the fovea) or can change from one person to another (e.g. the optic disk size ranges from 1 to 2 mm). Neither does the resolution of images give a satisfactory solution, even if it is the approach used in the literature. We propose to use the diameter of the field of view in images as a size invariant. It is shown to give good results, provided that all the images have been acquired with the same aperture angle. OPHDIAT is a telemedicine network for diabetic retinopathy screening. Thousands of color eye fundus images have been collected, 70% of which have been classified as healthy by ophthalmologists. The TeleOphta project aims at performing a preliminary analysis of the images, in order to automatically filter out healthy images, and thus reduce the burden on specialists. The proposed method has been validated using images from OPHDIAT. Its results have been compared with those of the spatial normalization based on the manual measurement of the distance between the center of the optic disk and the center of the fovea. Results show a nearly perfect agreement between them.