Internet censorship is usually reported as a yes/no: a site is blocked or it isn't. But how a country blocks is as telling as what it blocks — and the method leaves a fingerprint. Aggregating every blocking measurement by technique, the major censors look strikingly different from one another, and the differences line up with what the research literature has long documented.
| Country | n | Signature technique | TCP-reset | DNS | Interference | DPI |
|---|---|---|---|---|---|---|
| China | 8,213 | TCP-reset injection | 15% | 9% | 29% | 1% |
| Saudi Arabia | 1,815 | DNS blocking | 6% | 41% | 25% | 4% |
| Turkey | 3,076 | Diverse / multi-method | 11% | 24% | 15% | 11% |
| Russia | 10,275 | Connection interference | 2% | 7% | 34% | 4% |
| Iran | 4,752 | Generic / interference | 5% | 8% | 22% | 4% |
China stands out for TCP-reset injection (15% of its blocking measurements, three-to-seven times any other country here). This is the Great Firewall's textbook behavior: rather than dropping packets, on-path middleboxes forge a TCP RST to tear down the connection mid-flight. That it shows up so sharply in the data — on a base of 8,200+ measurements — is exactly what two decades of GFW research predicts.
Saudi Arabia is the mirror image: 41% DNS manipulation, by far the highest, reflecting a centralized DNS-based filtering regime rather than deep packet surgery. Turkey, meanwhile, has the most diverse toolkit of any country measured — DNS (24%), header manipulation (12%), TCP-reset (11%), and DPI/middlebox (11%) are all materially present. That breadth matches Turkey's well-documented history of escalating from DNS tampering (famously circumvented by writing Google's 8.8.8.8 on walls) to SNI/DPI filtering as users routed around each successive method.
Russia and Iran both lean on connection interference (34% and 22%) with comparatively little RST injection or DNS poisoning in the classified signal. Russia's profile is consistent with its ISP-level TSPU equipment throttling and degrading connections rather than cleanly resetting them.
These are tendencies, not a full accounting. A large share of blocking
measurements (39–52% for China, Russia, and Iran) carries no precisely
pinned method — that is method unresolved, not no block,
and it means the percentages describe the classified slice, not every
block. Sampling volume varies by country (South Korea, with only ~45 blocking
measurements, is excluded as too sparse to fingerprint). And technique mix
shifts over time as censors and circumventers co-evolve. Read this as a
qualitative fingerprint — the relative dominance of one method over
another — not a precise budget. The live breakdown for any country is at
GET /v1/measurement/techniques?country=XX.