We prove a convergence criterion for transformations to Poincaré-Dulac normal form that involves the centralizer of the given vector field. © 2000 Academic Press Since they were introduced by Poincaré and Dulac (and by Birkhoff for Hamiltonian systems), normal forms have proven to be a valuable tool in the local theory of ordinary differential equations. In many cases it turns out that a "partial" normal form (up to some finite degree in the Taylor expansion) is sufficient to provide an understanding of the structure of the solutions near a stationary point. Some problems, however, for instance the local analytic classification of vector fields, lead to the question whether a convergent transformation to normal form exists for a given vector field. It is known that often such a convergent transformation does not exist, and results by Bruno [2] indicate that convergence is indeed a rare phenomenon as soon as the normal form is not just equal to the linear part. (We will use this paper and the book [3] by Bruno as a basic reference.) In recent years, it has turned out that the question of convergence is closely related to the question of the existence of nontrivial "infinitesimal symmetries" for the given vector field. In fact, for two-dimensional systems it follows from results of Markhashov [11] and Bruno and Walcher [4] that there is a convergent transformation to normal form if and only if there is a nontrivial infinitesimal symmetry. Cicogna [5,6] extended parts of this result to higher dimension. Ito [9,10] proved convergence in the case of certain Hamiltonian systems under the assumption that there are sufficiently many integrals (in other words, sufficiently many canonical infinitesimal symmetries). For a survey of these and other results, see the paper [7] by Cicogna and Gaeta.