In this numerical study, a possible interface between experiment and simulation is developed. The main goal is to be able to interface experimental time histories, measured at some few locations, with the inlet section of a time varying numerical simulation of a spatially developing flow. The study is performed through two-dimensional DNS of a mixing layer by defining a virtual interface within the computational domain. First, necessary criteria to insure "realistic" inflow conditions for numerical simulations are analyzed. It is then shown that even if the Reynolds stresses and spectral distributions are properly taken into account, a correct representation of the space-time coherence of the flow organization is essential to obtain a good interface. Using the Linear Stochastic Estimation (LSE) of the whole inlet velocities from only three spatial monitoring locations, "realistic" downstream velocity fields can be generated. This realism is found to be valuable as well for turbulent statistics as for instantaneous snapshots of vorticity of the flow field. Two main applications of the present interface are then possible: experimental unsteady conditions can be used to drive a DNS, while the numerical simulation can be helpful as a tool for dynamical signal processing analysis of experimental data.