This claim is nonsensical.
The most common practice by far is to take an MD5 hash (or, increasingly, something else like a SHA hash), with a salt. The salt is a random string added to the plaintext password before it is hashed, and stored with the lot; it serves to make it hard to compute tables of all possible hashed passwords since this must then be done for each salt value.
Sometimes it is necessary to have the plaintext password, for some challenge-response algorithms. In this case the password may be unencrypted or it might be reversibly encrypted (using AES, DES, or something).
In both cases, the password is treated as a blob; the output of the algorithm is the only thing stored. I suggest, if you are interested, you hash a file; you may notice that the last character of the hash and the last character of the file are usually different.
Another way to use passwords for authentication is to use Kerberos, which involves using the password to generate a keypair; in this case the server only needs the public part of that keypair, and requires none of the characters of the password at all.
I can't think of a single widely-used password storage mechanism besides cleartext that stores a single character of the password in the clear.