Abstract. Numerical evidence is presented which strongly suggests that "Jacobi's last geometric statement"-that the conjugate locus from a point has exactly four cusps and the corresponding cut locus consists of only one topological segment-holds for compact real analytic Liouville surfaces diffeomorphic to S 2 if the Gaussian curvature is everywhere positive and has exactly six critical points, these being two saddles, two global minima, and two global maxima (as is the case for an ellipsoid). Our experiments suggest that this is a sufficient rather than a necessary condition. Furthermore, for compact real analytic Liouville surfaces diffeomorphic to S 2 upon which the Gaussian curvature can be negative but has exactly six critical points, these being two saddles, two global minima, and two global maxima, it appears that the cut locus is always a subarc of a line given by x 1 = const or x 2 = const, where (x 1 , x 2 ) are canonical coordinates with respect to which the metric has the form (f 1 (x 1 ) + f 2 (x 2 ))(dx 2 1 + dx 2 2 ). In the case of an ellipsoid, these curves are lines of curvature.The point of this paper is to present a conjecture, already contained in the abstract, as a contribution to pure mathematical research in global Riemannian geometry. The overwhelming bulk of the paper will however necessarily be devoted to a description of the computational methods used to motivate the conjecture, and, in particular, the checks which have been performed to verify that the software (which has been written specifically for this study) is indeed performing correctly. Such checks are vital in any experimental work. This paper is not intended to be a contribution to computational science as such, since the algorithms we have used, although perhaps combined in an unusual manner, are standard.The conjugate locus from a point p (which we will often refer to as the starting point) on a Riemannian manifold is the envelope of the geodesics emanating from p. In two dimensions, it can also be defined analogously to the nodes of a vibrating string as follows. A geodesic passing through p can be uniquely defined by the angle θ it makes at p with one fixed (reference) geodesic also passing through p. Varying this angle by a small amount (δθ) will cause the geodesic to deviate from its original path. The distance between points on the geodesic with angle θ (say g θ ) and the geodesic curve with angle θ + δθ (write g θ+δθ ) can be written as a real function ξ δθ of arclength s along g θ , where s = 0 corresponds to the point p. Clearly ξ δθ (0) = 0, since we are only considering geodesics which pass through p. The limit of ξ δθ (s)/δθ as δθ approaches zero may have further zeros (the nodes of the string, where, to