The recollection of particularly salient, surprising or consequential events is often called ‘flashbulb memories’. We tested people's autobiographical memory for details of 11 September 2001 by gathering a large national random sample (N = 678) of people's reports immediately following the attacks, and then by contacting them twice more, in September 2002 and August 2003. Three novel findings emerged. First, memory consistency did not vary as a function of demographic variables such as gender, geographical location, age or education. Second, memory consistency did not vary as a function of whether memory was tested before or after the 1-year anniversary of the event, suggesting that media coverage associated with the anniversary did not impact memory. Third, the conditional probability of consistent recollection in 2003 given consistent recollection in 2002 was p = .73. In contrast, the conditional probability of consistent recollection in 2003 given inconsistent recollection in 2002 was p = .18. Finally, and in agreement with several prior studies, confidence in memory far exceeded consistency in the long term. Also, those respondents who revealed evidence for consistent flashbulb memory experienced more anxiety in response to the event, and engaged in more covert rehearsal than respondents who did not reveal evidence for consistent flashbulb memory. Copyright © 2008 John Wiley & Sons, Ltd.