Treatment of coronary lesions by percutaneous transluminal angioplasty is performed from 2D observations acquired according to an optimal view, i.e. a point of view showing the segment including the stenosis in its most extended and unobstructed dimension over a given perimeter around the lesion. In clinical routine, the research of this view generally involves a large number of acquisitions with repeated injections of contrast agent. In this paper, we present an automatic method to determine this optimal view. We consider that a dynamic sequence of 3D coronary tree is available either from pre-segmented CT data or by reconstruction of the 3D coronary tree from the projections acquired at each phase of the cardiac cycle. We proceed by volume projection into the detector plane in all possible gantry orientations. The optimal viewing angle is the one that minimizes the degree of segment foreshortening and overlap with adjacent segments in projection space.