"Ferguson and Schneier, in their book Practical Cryptography, have argued the opposite: that MAC-then-encrypt (or MAC-and-encrypt) is the "natural" order and that encrypt-then-MAC is overly complex. The sore point of encrypt-then-MAC is that you have to be careful about what you MAC: you must not forget the IV, or (in case the protocol allows algorithm flexibility) the unambiguous identifier for the encryption algorithm; otherwise, the attacker could change either, inducing a plaintext alteration which would be undetected by the MAC. To prove their point, Ferguson and Schneier describe an attack over an instance of IPsec in which the encrypt-then-MAC was not done properly."
You generally have two options when it comes to authenticated encryption: use a specialized AEAD mode, in which the details of authentication are settled by the mode itself, or use "generic composition" --- encrypt securely, MAC securely, and safely combine the two operations. Specialized AEAD modes are preferable. But if you're going to do generic composition, the best current practice is encrypt-then-MAC.
Even if you encrypt-then-MAC, you can still forget to authenticate parameters (a good reason not to use generic composition). But if you MAC and then encrypt, you concede to attackers the ability to target the cipher's decryption operation directly with chosen-ciphertext attacks. Those attacks are powerful and have repeatedly broken TLS; they're also the most common form of attack on other cryptosystems (every padding oracle attack is a variant of them).
http://crypto.stackexchange.com/a/224