One of the most difficult problems to solve in rheology is prediction of the deformation of the melt during its “transient” conditions, i.e., before it reaches steady-state deformation. Most of our present understanding concerns the steady state when properties, e.g., viscosity or modulus, no longer depend on time and remain invariant. This article is concerned with the transient states obtained under non-linear conditions and asks why it sometimes takes a long time for the viscosity to reach steady state, when its terminal time τo calculated from the steady-state viscosity at that temperature, or from the crossover of G′ and G″ (τo = 1/ωx) indicates that the melt should relax in much shorter times. In pure viscometry (at constant shear rate), a transient behavior is expected before the melt reaches its steady state. Questions regarding the origin of this transient are experimentally and theoretically investigated and debated for a linear low-density polyethylene (M/M e ∼200). We show that the steady-state stability is a function of the melt's previous thermomechanical history. In order to study the instability of the melt under conditions that separate the effects of strain and strain rate, we investigate the triggering of time dependence of the moduli by strain in dynamic experiments performed in the non-linear range. We study, for linear polycarbonate and polystyrene, the effects of various parameters affecting the stability of entanglements under dynamic non-linear conditions, i.e., frequency, strain amplitude, and temperature. In all these experimental tests, we observe the same phenomenon: as strain is increased beyond a certain critical value (a function of frequency and temperature), the melt starts to become transient, e.g., its viscosity changes with time until it reaches a steady-state value. We review and rule out (for our results) the challenging interpretations based on melt edge fracture and melt surface decohesion. For transient and steady states obtained by pure viscometry or by dynamic shear oscillation, we compare the rheological properties [G′ (ω) and G″ (ω)] of the unsheared melt, before the transient, and of the “sheared melt,” after the transient, by performing a frequency sweep in the linear range. The results display large differences in G′ (ω) and G″ (ω) and in the value of the terminal time. In our proposed explanation we point to the limitation, as we enter the non-linear range, of the basic assumption of rheology regarding the scalability of the rheological parameters (stress and deformation tensors) in terms of viscosity, strain, and strain rate to describe the effect of the gap thickness. The definition of viscosity (ratio of stress and strain rate) and of strain rate (gradient of the velocity profile across the gap) requires a homogeneous melt, or, as we further propose, a homogeneous, unstructured entanglement network, which is a valid and justified concept only for certain conditions of deformation, e.g., in the linear viscoelastic range. A melt can be brought out of equilibrium with respect to its entanglement state. The return to equilibrium explains the transient properties. New entanglement states can be made quasistable, even at high temperature in the melt, by coupling entropic and enthalpic effects produced under specific conditions of melt processing. The currently accepted descriptions of rheology, we suggest, only apply to a stable entanglement state, which is not general enough and even becomes a severe limitation in the exploration of new frontier research in the application of rheology to processing of polymer melts. A supplemental table of contents is available for this article. Go to the publisher's online edition of the Journal of Macromolecular Science, Part B: Physics to view the free table of contents.