This paper studies a dynamical system that models the free recall dynamics of working memory. This model is an attractor neural network with n modules, named hypercolumns, and each module consists of m minicolumns. Under mild conditions on the connection weights between minicolumns, the authors investigate the long-term evolution behavior of the model, namely the existence and stability of equilibria and limit cycles. The authors also give a critical value in which Hopf bifurcation happens. Finally, the authors give a sufficient condition under which this model has a globally asymptotically stable equilibrium consisting of synchronized minicolumn states in each hypercolumn, which implies that in this case recalling is impossible. Numerical simulations are provided to illustrate the proposed theoretical results. Furthermore, a numerical example the authors give suggests that patterns can be stored in not only equilibria and limit cycles, but also strange attractors (or chaos).