In this study the efficacy of a quantitative analysis of backscattered ultrasound for the differential diagnosis of intraocular tumours is tested. The data acquisition was performed with an on-line acquisition and processing system for in vivo work. The backscattering spectra were corrected for the beam effects (diffraction and focussing). The measurements were taken from an animal model (rabbit) in which a Greene's amelanotic melanoma was implanted in the anterior chamber of the eye. The special histologic arrangement of this tumour makes it optimally suited to an assessment of the correlation between histology and backscattering cross section. Various backscattering models were considered theoretically and the choice for a practical testing was motivated on the properties of the observed backscattering spectra. We conclude that the backscattering model based on an "inhomogeneous continuum" with a cylindrical Gaussian autocorrelation function fits the data optimally. The relatively low correlation of the backscattering cross-sections to histology cannot yet be fully explained. The range of acoustic scatterer sizes, however, corresponds quite well to the dimension of observed and quantified histologic structures.