The formalism of low-rank recurrent neural networks is introduced. It is shown that trajectories of the dynamics of all N neurons stays in a D-dimensional space, if the connectivity matrix of of rank D. The relation to field models is highlighted.