A system and method for cleaning television viewing behavior data collected from set top boxes by detecting and correcting various problems that can occur in the viewing data. Three problems that may be detected and corrected by the system include: overflows of memory buffers; inclusion of non-human generated tuning events; and presence of clock offsets. After cleaning the television viewing behavior data, the cleaned data may be used to analyze audience viewing behavior in a manner that achieves a higher degree of accuracy than can be achieved by using uncleaned television viewing behavior data.