We employ VLTI GRAVITY to resolve, for the first time, the two images generated by a gravitational microlens. The measurements of the image separation \theta_{-,+}=3.78 +/- 0.05 mas, and hence the Einstein radius \theta_E =1.87 +/- 0.03 mas, are precise. This demonstrates the robustness of the method, provided that the source is bright enough for GRAVITY (K <~ 10.5) and the image separation is of order or larger than the fringe spacing. When \theta_E is combined with a measurement of the "microlens parallax" \pi_E, the two will together yield the lens mass and lens-source relative parallax and proper motion. Because the source parallax and proper motion are well measured by Gaia, this means that the lens characteristics will be fully determined, whether or not it proves to be luminous. This method can be a powerful probe of dark, isolated objects, which are otherwise quite difficult to identify, much less characterize. Our measurement contradicts Einstein's (1936) prediction that "the luminous circle i.e., microlensed image cannot be distinguished" from a star.