The precision of estimating endothelial cell density of the cornea, using a noncontact method of specular microscopy, was assessed by asking eight individuals with known densities in one operated-on and one unoperated-on eye to have photography on two occasions in the same day. Three photographs of each eye were assessed by two individuals using masking procedures. Interobserver differences were negligible and estimates were similar except for one eye with low density values. The method was applied to a natural history study in which 103 eyes had cell-density estimates at zero and two years, and mean cell loss was found to be 2%. A collateral study using only good photographs that were available at both zero and two years showed a similar loss of 1.86% and reduced the number of counts showing unexpectedly high gains and losses.