Once a lesson had been videotaped, the lesson was sent to a central processing unit where the videos were carefully tracked. The first step in the process was to check the quality of the videotaping to ensure that videographers followed the established filming procedures. Videotapes were then transcribed and translated using a state-of-the-art video processing software. Each transcription and translation was checked for accuracy. Each utterance or event in a lesson was then associated with a time point in the lesson to allow for analyses such as the duration of a lesson and the length of time spent reviewing homework. An international team of mathematics and education experts developed a set of codes to apply to the 1999 video data, initially based on the TIMSS 1995 Video Study but greatly expanded and improved. The codes were carefully reviewed by the national research coordinator in each participating country as well as a technical review panel assembled to assist with the study. The quality and reliability of the coding process was monitored carefully throughout the year-long process by calculating inter-rater reliability as well as the reliability of codes within and across countries. Coders were trained extensively before the coding process began, and were retrained when it was determined that they did not meet the minimum reliability established for the study. Coders were monitored carefully throughout the coding process to avoid reliability decay. Codes were dropped from the study if 85 percent reliability could not be achieved (or if individual coders could not reach at least 80 percent reliability). All codes presented in reports met or exceeded the minimum acceptable reliability standard established for the study.