Geometrical optics provides an excellent description for quasar images crossing caustics, which are formed by gravitational microlensing of objects like Q2237+0305. Within this approximation the source size can be estimated from the maximum magnification reached at caustic crossings. We evaluate the limitations imposed by diffraction on caustics using the formalism developed by Ulmer & Goodman. Close to a caustic a new characteristic length, smaller than the Fresnel length, enters the problem, limiting the angular resolution to about 200 femto arcsecond, or equivalently similar to 3 x 10(9) cm at the source. To achieve this resolution the brightness must be monitored at time intervals of a few seconds. If a significant fraction of quasar luminosity comes from sources smaller than those limits then interference effects would make the observed intensity oscillate, in a close analogy with a two-slit experiment. The characteristic period of such oscillations is expected to be about one-tenth of a minute. If such oscillations are detected, then photometry carried out at a single site may permit the determination of the caustic transverse velocity, and therefore may permit a direct conversion of the time units of brightness variations to the linear units at the source.