The Magiscan 2 Image Analysis System (M-2) was compared with the manual counting of asbestos fibers. The M-2 system consisted of a television camera connected to a standard phase contrast light microscope, a computer with keyboard and display screen, and a printer. The image of an 160 by 160 millimeter square of a sample slide was represented in memory by a 512 by 512 array of pixels with 64 levels of gray. Only the darkest areas of the image, representing particles, were detected for processing. Algorithms were provided to identify, screen, and analyze fibers. The instrument was evaluated using 178 field samples. Manual counts were done by the NIOSH Physical and Chemical Analysis Method 239. One set of 72 slides was counted by the M-2 and then sent out to 12 laboratories participating in the Proficiency Analytical Testing Program. Data were transformed to a log scale to give a constant coefficient of variation (CV) over the entire range of measurements. Intralaboratory and interlaboratory CVs were 0.23 and 0.46 at the low concentration and 0.26 and 0.43 at the high concentration, respectively. The Magiscan CV fell in the range of estimates for intracounter variability. Problems encountered with M-2 included not counting broken and large fibers, counting chains of fibers as one fiber, and missing fibers at higher concentrations. The authors conclude that the M-2 system is at least as precise as manual counters, is less prone to interlaboratory bias, is somewhat faster than manual counters, reduces operator fatigue, and requires less operator expertise.