The phase transition in the mean-field XY model is shown analytically to be related to a topological change in its configuration space. Such a topology change is completely described by means of Morse theory allowing a computation of the Euler characteristic-of suitable submanifolds of configuration space-which shows a sharp discontinuity at the phase transition point, also at finite N . The present analytic result provides, with previous work, a new key to a possible connection of topological changes in configuration space as the origin of phase transitions in a variety of systems. PACS number(s): 05.70.Fh; 75.10.Hk Phase transitions (PTs) are one the most striking phenomena in nature. They involve sudden qualitative physical changes, accompanied by sudden changes in the thermodynamic quantities measured in experiments. From a mathematical point of view, both qualitative and quantitative changes at PTs are conventionally described by the loss of analyticity of the probability measures and of the thermodynamic functions. According to statistical mechanics, in the grandcanonical and canonical ensembles, such a nonanalytic behavior can exist only in the thermodynamic limit, i.e., in the rather idealized case of a system with N → ∞ degrees of freedom [1]. PTs in real systems would then be the "shadow", at finite but large N , of this idealized behavior. The way nonanalytic behavior can emerge as N → ∞ due to singularities of the probability measures of the statistical ensembles has been rigorously studied by Yang and Lee in the grandcanonical ensemble [2] and by Ruelle, Sinai and others [3] in the canonical ensemble; however, to the best of our knowledge, no equally rigorous approaches to this problem exist in the microcanonical ensemble [4]. Moreover, the necessity of taking the N → ∞ limit to speak of PTs seems less satisfactory today, since there is growing experimental evidence of PT phenomena in systems with small N (e.g., atomic clusters and nuclei, polymers and proteins, nano and mesoscopic systems). There is also a deeper reason why the conventional approach to PTs is not yet completely satisfactory. Consider a classical system described by a Hamiltoniani is the kinetic energy, V (ϕ) is the potential energy and ϕ ≡ {ϕ i } and π ≡ {π i }'s are, respectively, the canonical conjugate coordinates and momenta. Although in principle all the information on the statistical properties is contained in the function V (ϕ), no general result is available to specify which features of V (ϕ) are necessary and sufficient to entail the existence of a PT. This is the more surprising since in many cases, knowing a priori that a system undergoes a PT, several relevant properties of the PT can be predicted just in terms of very general features of V (ϕ) (e.g., by means of renormalization-group techniques). In the light of some recent results [5][6][7][8][9][10][11] an alternative new approach approach seems actually possible by resorting to topological concepts: PTs would then be related to topology changes (TCs) of suitable ...