Because of the diagnostic and prognostic value of gastroesophageal endoscopic signs in cirrhosis, we studied their interobserver and intercenter agreement. Fifty cirrhotic patients without previous digestive bleeding were included in this study. The gastroesophageal endoscopic examinations of these patients, recorded on video tape, were independently viewed by 12 observers belonging to different hospitals. Agreement was evaluated by the kappa index (k), the values of which were given with p less than 0.001. Agreement was good for the presence (k = 0.40) or the size (k = 0.51) of esophageal varices as well as for red signs on the mucosa overlying esophageal varices, but it was poor for the color (k = 0.10) or the extent (k = 0.20) of esophageal varices. Agreement was poor for all gastric signs: fundic varices (k = 0.35), red spots (k = 0.28), associated lesions (k = 0.27), mosaic pattern (k = 0.27), gastropathy (k = 0.11). The mean diameter of esophageal varices, evaluated without any scale, varied from 3.5 +/- 1.4 to 7.5 +/- 4.0 mm between the 12 observers (p less than 0.001). Agreement was maximum for the more developed grade of a sign and minimum for the intermediate grade. The value of kappa indexes in this intercenter study was compared to that obtained in a previous study from one institution: interobserver agreement was significantly better within one center than between different centers. In conclusion, intercenter agreement is good for the size of esophageal varices and the presence of red signs, and poor for all the other signs. The millemetric measurement of esophageal varices diameter, without any scale, appears to be unreliable.(ABSTRACT TRUNCATED AT 250 WORDS)