We discuss an attractor neural network where both neurons and synapses
obey dynamical equations; it models the compresence of permanent and
short-term memories. By signal-noise analysis and a discussion in term
s of an energy-like function, we outline the behavior of the model.