Simulations. (A and B) Estimation error of the closing angle θ (N = 1,800 sites). (A) The comparison of fitted θ of simulated clathrin coat structures to corresponding ground-truth θ shows no systematic bias and a narrow spread of the error across the ground truth although the spread increases when θ < 20° or θ > 160°. We reasoned that in the earlier range, where the structures are flat, the slightly increased error corresponds to the insensitivity of θ to flat structures. In the later range, where the vesicles are almost closed, the error was caused by indistinguishable tiny holes corresponding to the real vesicle openings and unlabeled clathrin in the coat. The fitted structures were simulated to have similar quality as the experimental data and to distribute evenly across θ. (B) The distribution of θ shows no significant error compared to the expectation corresponding to the evenly distributed θ, except for the small potential underestimation of entirely closed coats. (C–G) Averaging preserves U-shapes in simulated endocytic sites. (C) A ground-truth structure was used for simulating U-shaped clathrin coats. The model for simulation was built by combining a hemisphere with a radius of R = 97 nm and a cylinder with a height D = 0.5 R. We choose the radius value according to the median radius of bin 6 in Fig. 4 G. This bin has a closing angle slightly larger than 90°. 20-nm-thick cross-sections of the averages (N = 100 sites) registered based on (D) the ground truth and (E) the spherical fit are shown with the ground-truth U-shaped model (dotted line). Histograms of normalized estimation errors are shown for the parameters (F) surface area and (G) radius, with mean values of −2.6 and 2.5% respectively. The scale bar is 100 nm.