We discuss the theoretical foundations for testing nonlinear vacuum electrodynamics with Michelson interferometry. Apart from some nondegeneracy conditions to be imposed, our discussion applies to all nonlinear electrodynamical theories of the Plebański class, i.e., to all Lagrangians that depend only on the two Lorentz-invariant scalars quadratic in the field strength. The main idea of the experiment proposed here is to use the fact that, according to nonlinear electrodynamics, the phase velocity of light should depend on the strength and on the direction of an electromagnetic background field. There are two possible experimental setups for testing this prediction with Michelson interferometry. The first possibility is to apply a strong electromagnetic field to the beam in one arm of the interferometer and to compare the situation where the field is switched on with the situation where it is switched off. The second possibility is to place the whole interferometer in a strong electromagnetic field and to rotate it. If an electromagnetic field is placed in one arm, the interferometer could have the size of a gravitational wave detector, i.e., an arm length of several hundred meters. If the whole interferometer is placed in an electromagnetic field, one would have to do the experiment with a tabletop interferometer. As an alternative to a traditional Michelson interferometer, one could use a pair of optical resonators that are not bigger than a few centimeters. Then the whole apparatus would be placed in the background field and one would either compare the situation where the field is switched on with the situation where it is switched off or one would rotate the apparatus with the field kept switched on. We derive the theoretical foundations for these types of experiments, in the context of an unspecified nonlinear electrodynamics of the Plebański class, and we discuss their feasibility. A null result of the experiment would place bounds on the parameters of the theory. We specify the general results to some particular theories of the Plebański class; in particular, we give numerical estimates for Born, Born-Infeld and Heisenberg-Euler theories.