Geographic and temporal trends of rotavirus detections in the United States for the period January 1989-May 1991 were determined by analyzing data reported monthly by 47 virology laboratories participating in the North American Rotavirus Surveillance System. Reports included complete information on the number of specimens tested, the number of test results positive for rotavirus, and the method used to detect rotavirus. Consistent trends in regional and geographic area were identified, with distinctly different peaks of rotavirus activity in the western and eastern states. Each year in the western states, rotavirus activity began in November and peaked in December-January, whereas in the eastern states activity began in January and peaked in February-March. These differences do not correlate with obvious trends in strain variation of rotavirus and remain unexplained. Unexpected reporting of summer rotavirus activity by some laboratories in 1989 was traced to the use of a single diagnostic kit and to two questionable laboratory practices: having more than six medical technologists perform the test and failure to use controls with the test. Laboratory-based surveillance of rotavirus activity has proven to be useful in identifying and correcting problems in laboratory methods for detecting rotavirus and will be a sensitive means for monitoring coverage of the rotavirus vaccine now being developed.