In order to connect to TCP, a hacker must be able to see the responses. This is because the server will send the hacker its "Initial Sequence Number (ISN)", which must used in all of the subsequent packets sent to the server. Therefore, blind IP spoofing will not work with TCP, in theory.
The problem is that many machines use predictable ISNs. Therefore, a hacker can connect to the machine, find the current ISN, then predict what the ISN will likely be in the subsequent connection.
For example, some systems simply add 64k to the ISN of a previous connection, so a hacker can connect once, then add 64k to the spoofed connection.
