We address inference of dynamic metabolic network models from steady-state metabolite concentration and flux data. As a case study we rely on simulation of an example branched metabolic pathway. In general, steady-state data fitting does not guarantee appropriate dynamical performance. In order to tackle this problem, we propose tools for the assessment of the accuracy of dynamical model approximations, and suggest the use of constraints on the expected dynamical system behavior, in the form of reaction-rate constraints over a region of interest, to improve the dynamical properties of the model inferred. Numerical results demonstrate the potential of the approach.