An analysis of three-dimensional ultrastructure is especially efficient when performed on stereo images. The high voltage electron microscope (HVEM) providing high beam penetration and low radiation damage to specimens has been shown to be the most effective instrument for such studies. The successive HVEM stereo images recorded on films are subsequently photoprinted or stored in an image processor memory can be observed on an RGB monitor. Successful stereo imaging must meet several requirements regarding magnification, specimen thickness, tilt axis and angle. In particular, the optimal tilt for stereo perception depends on the thickness and microscopical magnification. Ultramicrotome settings can provide only approximate information on section thickness; however, their sectioning compression, procedural etching or beam induced shrinkage has to be considered. The situation becomes even more difficult when extractable media are used because an amount of extracted material may seriously affect final thickness. An estimation of an exact thickness of whole cell mounts is practically impossible, and therefore, optimal tilt difficult to predict. Then the most reliable way is the laborious recording of series of images at different tilt angles and selection of the most effective. To overcome this constraint, I designed a very reliable system for AEI HVEM 7MkII allowing exact prediction of the perfect tilt angle. The diagram of the system is presented on Fig. 1 and a photograph of the assembly on Fig. 2. It consists of a low light level TV camera with two independent outputs-SIT/TVC/, image processor-Quantex/QIP/, two high resolution flat square monitors-Hitachi/H1 & H2/ and stereo viewer-Nikon/SN/. A signal from the first TV camera output goes to the QIP, after processing is displayed on monitor H1. A signal from the second output goes directly to the second monitor, H2. Prerequisites for using this system are axis-centered or eucentric goniometer and a rotation-free imaging system of HVEM. The recording procedure happens as follows. The specimen is tilted 1/2 of the value of the roughly predicted optimal angle and the resulting image is stored in the Quantex memory followed by recording on the film sheet. The left monitor displays the stored image from QIP, while the right monitor provides the live image. Then the specimen is tilted while being observed with a stereo viewer. When the strongest stereo effect is noticed, then tilting is stopped and an image recorded. If + and − tilt values differ, then perfectly symmetrical values can be calculated followed by image recording. Figs. 3 and 4 represent the stereo images of a human melanoma cell grown on a film grid. An optimal stereo perception was obtained for different magnifications, at different tilt angles determined using the system I have designed.