Telepathology affords the means to provide pathological diagnosis and consultation to remote sites. However, before telepathology can become an acceptable medical tool, it will be vital to determine the diagnostic accuracy of this technology. We report the results of a single-blind study of the accuracy of diagnosis performed using computerized still images obtained from a telepathology workstation used in a French telepathology network. Four pathologists, each working alone, reviewed a total of 200 cases of routine surgical pathology (50 cases each), and performed diagnosis based on both computer CD-ROM still images (CD) and conventional glass slides (GS). Concordance between GS and CD diagnosis, as well as accuracy, were determined. Other factors related to performance were also measured, including diagnostic certainty, reasons for uncertainty, and causes of diagnostic error. Overall, there was good agreement between CS and CD diagnosis. There was 87.5% concordance between CS and CD diagnosis, and comparison to consensus (correct) diagnosis showed accuracy of 95.5% and 88.5% for GS and CD diagnosis, respectively. Marked variability in accuracy of CD diagnosis was observed among the four pathologists, and issues related to image selection and/or quality appeared responsible for 60% of the diagnostic errors. The lack of sufficient images and clinical information were frequently cited as reasons for diagnostic uncertainty, as well as feelings of insufficient expertise. It is likely that the opportunity for interaction with the referring pathologist and the use of subspecialty consultants would likely improve the performance of telepathology.