The variation of the circulus arteriosus is studied using multivariate methods. The data which form the basis of this study are 19 measurements of half the circumference of the arteries that form the circle of Willis and its afferent and efferent branches; 100 circles of Willis were measured for this purpose. Since the number of variables per individual is large, multivariate statistical techniques are the most appropriate method to gain insight in the relations of vessel sizes that exist within the circle of Willis. So a principal component analysis was performed on the data. The results clearly show a number of relations between vessel sizes. In general, inverse relationships were found of vessels that have (at least partially) an identical irrigation area: both internal carotid arteries and the ipsilateral posterior communicating artery show an intimate relationship and are together inversely related to the basilar artery and the precommunicating part of the posterior cerebral artery. Inverse relationships are also found for both vertebral arteries and both precommunicating parts of the anterior cerebral arteries. The homonymous efferent arteries appear to be closely related and show an independent variation. Together the first six principal components explain 69% of the variance. These results support a haemodynamical hypothesis on the explanation of the variability of the circle of Willis. Moreover, the differential growth in the head-neck region during the first two decades of life is postulated to be the origin of a part of the variation.