By ultrathin serial sectioning of frog mesenteric capillaries it was recently demonstrated that the many apparently free vesicles in electron microscope (EM) sections of endothelial cells may be artefacts due to conventional (500-700 A thick) sectioning (Frøkjaer-Jensen, 1980). The vesicles were found to be part of two sets of invaginations of the cell surfaces; one set connected to the lumen, the other to the interstitium. The present study extends this view to comprise the vesicle organization in frog striated muscle capillaries. By analysis of the three-dimensional organization of the plasmalemmal vesicles in 21 ultrathin serial sections (120-150 A) of two muscle capillaries it is demonstrated that less than 1% of the about 70% apparently free vesicles seen in conventional thin sections of the same capillaries in fact represent truly free vesicular units. By analysis of 15 conventional EM cross-sections of capillaries from the frog cutaneous-pectoris muscle containing plasmaproteins in high concentration it is furthermore demonstrated that 48% of the total vesicle population connect to the lumen at the time of fixation. This organization of the vesicular system seems incompatible with the concept that macromolecules are transferred across the capillary wall by vesicular transport or by a series of fusions and fissions between individual cytoplasmic vesicles but is compatible with the notion that macromolecules exchange across capillary walls by means of passive processes such as diffusion and convection through rare 'large pores'. The study emphasizes that any attempts to classify vesicles in conventional thin sections as 'luminal', 'cytoplasmic' and 'abluminal' is impossible and may lead to erroneous interpretations of vesicle involvement in transcapillary exchange of macromolecules. The rare occurrence of transendothelial channels compared to the number of vesicle invaginations suggests that the main function of the vesicular system relates to functions other than transport.