The electrophoresis of long polyelectrolytes is considered theoretically, with special attention to duplex DNA. We first discuss quantitative approaches to determine unambiguously the entanglement properties of polymer solutions. Following an idea proposed by Grossman and Soane, we then assume that the "mesh" size of the solution plays the role of a dynamic "pore size" in order to apply theories for gel electrophoresis. In the framework of the Ogston model, we predict that duplex DNA up to 1 kb or more should be separable in dilute (i.e. nonentangled) solutions of high molecular weight polymers. In an entangled solution, and for DNA larger than the pore size, we use a recently developed fluctuation-reptation model to predict the range of sizes in which separation should be possible as a function of electric field E and pore size zeta b. For zeta b larger than the Kuhn length of DNA, we predict a separation up to a size N*scaling as E-1 zeta b-1. For zeta b smaller than the Kuhn length, two different regimes are expected. For small electric fields (typically of the order of 10 V/cm), N*should be proportional to E-1 zeta b-3/2, whereas for high electric fields such as encountered in capillary electrophoresis, we expect that N*is proportional to E-2/5 zeta b-12/5. These predictions are qualitatively different from earlier ones. Finally, we demonstrate that the finite lifetime of the "pores" in an entangled solution (as opposed to a gel) may lead to a new migration mechanism by constraint release, which is not size-dependent.(ABSTRACT TRUNCATED AT 250 WORDS)