The purpose of the study was to examine intercenter variability in the interpretation of Gram-stained vaginal smears from pregnant women. The intercenter reliability of individual morphotypes identified on the vaginal smear was evaluated by comparing them with those obtained at a standard center. A new scoring system that uses the most reliable morphotypes from the vaginal smear was proposed for diagnosing bacterial vaginosis. This scoring system was compared with the Spiegel criteria for diagnosing bacterial vaginosis. The scoring system (0 to 10) was described as a weighted combination of the following morphotypes: lactobacilli, Gardnerella vaginalis or bacteroides (small gram-variable rods or gram-negative rods), and curved gram-variable rods. By using the Spearman rank correlation to determine intercenter variability, gram-positive cocci had poor agreement (0.23); lactobacilli (0.65), G. vaginalis (0.69), and bacteroides (0.57) had moderate agreement; and small (0.74) and curved (0.85) gram-variable rods had good agreement. The reliability of the 0 to 10 scoring system was maximized by not using gram-positive cocci, combining G. vaginalis and bacteroides morphotypes, and weighting more heavily curved gram-variable rods. For comparison with the Spiegel criteria, a score of 7 or higher was considered indicative of bacterial vaginosis. The standardized score had improved intercenter reliability (r = 0.82) compared with the Spiegel criteria (r = 0.61). The standardized score also facilitates future research concerning bacterial vaginosis because it provides gradations of the disturbance of vaginal flora which may be associated with different levels of risk for pregnancy complications.