Correct. It depends on how well they separated those transactions. All, some or none of their previous transactions could be linked. The large coin transactions that they analyzed didn't go through the mixing services that exist today. It was much more crude. They were probably manually broken up into many addresses and rejoined. The better mixing services today use separate pools of addresses, so there's zero taint or linkage between them. A temporary ledger that keeps account of which coins go to which addresses is maintained in a separate database, and supposedly they delete that info after 6 confirmations. However, it could still be possible to link transactions in separate address pools, assuming you can identify the addresses in them. If you know that a certain mixing service charges a 1.5% fee, and you see 100 bitcoins go in (in any number of transactions, from/to any number of addresses), then in the other pool you see 98.5 bitcoins come out (ditto), you could link the two. This is difficult with large mixing services and many transactions, but theoretically possible. The defense is to buy more bitcoins than you need. If you need 30 BTC, buy 50, send them to the mixing service and then send 30 to SR and 20 somewhere else. That makes the coins much harder to link. Not all, because there isn't good data on a lot of purchases. Consider the people who use cash drops or buy through the mail. The addresses they use will always be hard if not impossible to link to an identity, and that's a matter of traditional detective work that no statistical analysis on the block chain can solve.