Furthermore, the sensitivity analysis demonstrated that the Google Flu AE9C90CB Trends correlation with the reference standard was more influenced by outlier observations than was CDC ILI Surveillance data. Most of the influential observations occurred during the peak 2003�C04 influenza season. This season was characterized by early and intense influenza activity, a large number of influenzaassociated pediatric deaths, and increased media attention to influenza. It is possible that during this influenza season, physician laboratory testing patterns or patient health care seeking behavior differentially affected the relationship between ILI rates and laboratory confirmation of influenza. Additionally, internet search behavior about respiratory infections during this period could have been different than during subsequent, more typical influenza seasons. These findings are relevant to the applicability of surveillance using internet key word searches during the 2009 H1N1 pandemic and future anomalous influenza seasons. The Google Flu Trends statistical model was created and validated using rates of ILI, which is a nonspecific syndrome that is not necessarily caused by influenza virus infection, but used for decades as an indicator of the burden of outpatient influenza illness. Any nationwide surveillance using internet key word search is likely to be most representative if the search engines being monitored has widespread use. As the popularity of a particular internet search engine wanes, so too may the overall accuracy of disease activity estimates using its data. Also, the stability of internet key word surveillance relies on consistency of internet search behavior, as well as search term use between geographic regions and over time. Changing media trends, word search choices, and cultural make-up of regions and over time may also affect the representativeness of internet search surveillance. Our analyses represent the first comparison of Google Flu Trends data with data on DMPQ dihydrochloride laboratory-confirmed influenza virus infections. Prior studies have demonstrated that Google Flu Trends can estimate rates of nonspecific ILI in New Zealand, Europe, and the United States.