A recent method for the global optimization of a real valued function defined in the $n$-dimensional euclidean space has been applied to the detection of the most stable conformation of a molecule. The method, inspired by the quantum mechanical theory of diffusion across potential barriers, finds the deepest minimum of the intramolecular energy hypersurface by following the asymptotic behaviour of the paths of a system of stochastic differential equations. As an example of its application, the search for the global minimum of the deoxyribose-phosphate-deoxyribose fragment (regarded as the smallest unit representative of polynucleotide backbones) is described.