In order to use a gravitational lens to measure the Hubble constant accurately, it is necessary to derive a reliable model of the lens surface potential. If the analysis is restricted to the locations and magnifications of point images, the derived Hubble constant depends upon the class of mass models used to fit the data. However, when there is extended emission from an Einstein ring, it may be possible to derive a potential from the observed surface brightness in a model-independent manner. This procedure is illustrated with reference to B1608+656. The multi-band images are de-reddened, de-convolved and de-contaminated so that the luminous matter and the surface brightness contours in the Einstein ring are both faithfully mapped. This intensity distribution can then be used to reconstruct the potential. Progress in implementing this program is reported. The observed incidence of multiple-imaged galaxies on the Hubble Deep Fields is an order of magnitude smaller than naively predicted on the basis of radio lens surveys, like CLASS, but consistent with the rate computed using surface photometry of candidate lens galaxies assuming standard mass to light ratios. In order to resolve this paradox, it is suggested that most galaxy lenses are located in compact groups.