Two particles, initially in a product state, become entangled when they come together and start to interact. Using semiclassical methods, we calculate the time evolution of the corresponding reduced density matrix ρ1, obtained by integrating out the degrees of freedom of one of the particles. To quantify the generation of entanglement, we calculate the purity P(t) = Tr[ρ1 (t) 2 ]. We find that entanglement generation sensitively depends (i) on the interaction potential, especially on its strength and range, and (ii) on the nature of the underlying classical dynamics. Under general statistical assumptions, and for short-ranged interaction potentials, we find that P(t) decays exponentially fast if the two particles are required to interact in a chaotic environment, whereas it decays only algebraically in a regular system. In the chaotic case, the decay rate is given by the golden rule spreading of one-particle states due to the two-particle coupling, but cannot exceed the system's Lyapunov exponent. When two systems (. . . ) enter into temporary interaction (. . . ), and when after a time of mutual influence the systems separate again, then they can no longer be described in the same way as before, viz. by endowing each of them with a representative of its own. This is how entanglement was characterized by Schrödinger almost seventy years ago [1]. Entanglement is arguably the most puzzling property of multipartite quantum systems, and often leads to counterintuitive predictions due to, in Einstein's words, spooky action at a distance. Entanglement has received a renewed, intense interest in recent years in the context of quantum information theory [2].In the spirit of Schrödinger's above formulation, one is naturally led to ask the following question: "What determines the rate of entanglement production in a dynamical system ?". Is this rate mostly determined by the interaction between two, initially unentangled particles, or does it depend on the underlying classical dynamics ? This is the question we address in this paper. Previous attempts to answer this question have mostly focused on numerical investigations [3,4,5,6,7,8], with claims that entanglement is favored by classical chaos, both in the rate it is generated [3,4,7] and in the maximal amount it can reach [5]. In particular, strong numerical evidences have been given by Miller and Sarkar for an entanglement production rate given by the system's Lyapunov exponents [4]. These findings have however been recently challenged by Tanaka et al. [6], whose numerical findings show no increase of the entanglement production rate upon increase of the Lyapunov exponents in the strongly chaotic but weakly coupled regime, in agreement with their analytical calculations relating the rate of entanglement production to classical time correlators. Ref.[6] is seemingly in a paradoxical disagreement with the almost identical analytical approach of Ref. [7], where entanglement production was found to be faster in chaotic systems than in regular ones [9]. This controversy t...