We discuss the performance of a next generation νμ( νe)↔ ντ oscillation experiment, based on a target consisting of a sandwich of nuclear emulsion layers and silicon detectors. The target would be followed by a full spectrometer for energy-momentum measurement and particle identification. A τ-enriched sample is selected in terms of vertex and kinematical criteria using the electronic spectrometer, with very high efficiency, while at the same time reducing the load of background events to be scanned by up to two orders of magnitude. Events fulfilling the selection criteria are very efficiently located and scanned in the emulsion, thanks to the very good resolution allowed by the silicon tracker. In terms of the two-neutrino families mixing angle, the sensitivity expected by the CHORUS experiment at CERN is sin 2 2 θμτ < 2.8 × 10 -4 while the proposed experiment COSMOS at FNAL has a sensitivity which is predicted to be an order of magnitude better. We present Monte Carlo calculations showing that a detector based on an emulsion-silicon target can reach a sensitivity up to sin 2 2 θμτ < 1.1 × 10 -5 in an extended run at CERN and up to sin 2 2 θμτ < 3.2 × 10 -6 at FNAL.