Semi-quantitative diagnostic features were extracted by a visual analysis of the echographic images of selected cases of breast disease and the results stored in a computer database. The long term aim is to create an environment suitable for the use of multivariate statistical methods systematically to evaluate ultrasound interpretive criteria and diagnostic performance in relation to factors such as scanning instrumentation and other diagnostic techniques. Eventually it is hoped that it will be possible to generate a system for computer assisted diagnosis and training. The results of this pilot study serve to demonstrate the feasibility of the approach and a univariate analysis is used to provide a preliminary ranking of diagnostic features. Features found to be particularly valuable for distinguishing benign from malignant solid lesions were the regularity and definition of the edge of the tumour, the mobility of the tumour and measures of echo heterogeneity within and posterior to the tumour mass.