Large mirrors appear to be limited in resolution by their medium scale defects (mirror surface "ripples" of small amplitude and of 4 to 12 cycles per diameter). The ultraviolet resolutions of the Pic du Midi (PDM) 2 m mirror, the Canada-France-Hawaii (CFH) 3.6 m mirror, and the Space Telescope (ST) 2.4 m mirror have been studied. Hartmann screen tests were used to determine the wavefront error of the PDM and CFH mirrors, while only semiquantitative reports were used in the case of the ST mirror. Point spread function and modulation transfer function were determined by two-dimensional fast Fourier transform analysis, a necessary technique in the ultraviolet, where the small phase defects of mirrors are no longer negligible (and also because the mirrors' irregularities do not possess a particular arrangement or a given symmetry). If the three mirrors appear to be nearly limited by diffraction in the visible (except the CFH), in the ultraviolet at Lyman a 121.6 nm the resolution is inferior by more than a factor of 10 to the diffraction-limited resolution. While the ultra-violet wavelength range and very high resolution (0.014 arcsec with a diffrac-tion-limited Space Telescope at Lyman a) are highly desirable, such a limitation imposed by mirror surface quality has to be mentioned. The effects of the correlation length, amplitude of defects, and Hartmann screen sampling are presented, as well as some comments on the inadequacy in the ultraviolet of analytical methods compared to two-dimensional numerical simulations.