Internet Filtering in China in 2004 2005 

requested page did not exist (known as a  404 error ),

 152

 even though we could successfully, and 

simultaneously, access the page from outside China.  Furthermore, our testing of  control  sites known to 

be accessible in China (for instance, Web sites of state agencies) through proxy servers returned errors for 

approximately 20% of our requests.  In our proxy testing analysis, we evaluate any site that we could 

access from outside China, but not from inside China in the majority of our tests, as blocked.  We also 

attempted to identify overblocking, where China's filtering prevented access to pages with unrelated 

content at domain names and URLs similar to those of sites identified by news reports and previous 

testing as containing sensitive content. 

Moreover, China's filtering system presents a risk of  false positive  results during testing.  When 

a user attempts to access a blocked site, the filtering system resets the user's connection to that site using 

a TCP RST packet; subsequently, the system advertises a TCP window size of zero for that Web server.  

The user is unable to connect to that server's IP address until the system advertises a window size greater 

than zero.  Our in state testing included an instance of this condition: a domain that had been accessible 

prior to the testing of a URL (www.poets4peace.com/peacehall.htm) containing a potentially targeted 

string was blocked after a request for a URL on that site with a prohibited keyword.  Even though 

subsequent URL requests were for unrelated content, the disruption caused by the zero window size 

condition persisted and prevented the tester from reaching those pages. 

We conducted testing in several stages, and analyze results by topic (below). The first component 

was similar to our  high impact  list tests in other states we examined, checking a collection of URLs with 

content on or domain names including sensitive topics (for example, a page with Falun Gong material, or 

a URL such as www.falungong.com).  The second testing phase built on our past data on filtering in China 

by checking a set of domains known to be of concern.  Additionally, we tested URLs containing words or 

strings of letters similar to those found in pages on blocked topics (for example, testing URLs containing 

the strings  falu  or  flg  to probe filtering of Falun Gong content), but that hosted pages with content 

unrelated to the sensitive subject.  Finally, we tested a  long list  comprised of the top Google search 

results, in English and in Chinese, for keywords known to be sensitive and filtered.

153

D. Comparison of Testing Methods 

Due to the complications of testing filtering in China, we sought to combine both proxy testing 

and in state testing for this report.  We found a 78% correlation between the two methods, with almost all 

the sites we were able to access through the proxy also accessible in state.  However, only 60% of the sites 

identified as blocked during proxy testing were confirmed as such during the in state portion.  The 

discrepancies were concentrated in the sites tested to determine overblocking (filtering of URLs similar to 

those containing targeted content, but with unrelated subject matter):  82% of the blocks during proxy 

testing of sites sensitive content were confirmed during in state testing and only 22% of the 

corresponding blocks of unrelated content were thus verified.  Potential explanations include the difficulty 

152

 See 404 error, Wikipedia, at http://en.wikipedia.org/wiki/404_error. 

153

 ONI compiled these keywords from a list of terms blocked by the instant messaging software QQ.  See The Words 

You Never See in Chinese Cyberspace, China Digital News, at 

http://journalism.berkeley.edu/projects/chinadn/en/archives/002885.html#more (Aug. 30, 2004). 

22