The quantification of axon diameter from electron microscopy (EM) images requires a choice of heuristics to approximate the dimensions. This choice impacts the resulting size distributions and should be adapted to the task. In diffusion MRI, complex shapes are encoded based on their directional mean squared displacement. We compute diffusion-specific ground truth diameters using Monte-Carlo simulations and evaluate typical choices of EM image analysis heuristics against them. Further, we propose a new and simple heuristic based on the mean square displacement inside an ellipse to compute better axon diameter quantifications from electron microscopy.