||This article includes a list of references, but its sources remain unclear because it has insufficient inline citations. (November 2010)|
A new salt is randomly generated for each password. In a typical setting, the salt and the password are concatenated and processed with a cryptographic hash function, and the resulting output (but not the original password) is stored with the salt in a database. This allows for later authentication while defending against compromise of the plaintext password, even in the event that the database is somehow compromised. The original intent of salting was primarily to defeat pre-computed rainbow table attacks that could otherwise be used to greatly improve the efficiency of cracking the hashed password database. A greater benefit now is to slow down parallel operations that compare the hash of a password guess against many password hashes at once.
Salts are closely related to the concept of nonce.
Unix implementations 
Earlier versions of Unix used a password file (
/etc/passwd) to store the hashes of salted passwords (passwords prefixed with two-character random salts). In these older versions of Unix, the salt was also stored in the passwd file (as cleartext) together with the hash of the salted password. The password file was publicly readable for all users of the system. This was necessary so that user-privileged software tools could find user names and other information. The security of passwords is therefore protected only by the one-way functions (enciphering or hashing) used for the purpose.
Early Unix implementations limited passwords to 8 characters and used a 12-bit salt, which allowed for 4,096 possible salt values. While 12 bits was sufficient for the 1970s, by 2005 disk storage had become inexpensive; so much so that an attacker could pre-compute the hashes of millions of common passwords, including all 4,096 possible salt variations for each password, and store the precomputed values on a single hard drive. An attacker with a larger budget could build a disk farm with all 6-character passwords and the most common 7- and 8-character passwords stored in hashed form for all 4,096 possible salt values.
Web application implementations 
It is common for a web application to store in a database the hash value of a user's password. Without a salt, a successful SQL injection attack may yield easily crackable passwords. Because many users re-use passwords for multiple sites, the use of a salt is an important component of overall web application security. Some additional references for using a salt to secure password hashes in specific languages (PHP, .NET, etc.) can be found in the external links section below.
A public salt makes it more time-consuming to crack a list of passwords. However, it does not make dictionary attacks harder when cracking a single password. The attacker has access to both the encrypted password and the salt, so when running the dictionary attack, the attacker can simply use the known salt when attempting to crack the password.
To understand the difference between cracking a single password and a set of them, consider a single password file that contains hundreds of usernames and passwords. Without a salt, an attacker could compute hash(attempt), and then check whether that hash appears anywhere in the file. The likelihood of a match, i.e. cracking one of the passwords with that attempt, increases with the number of passwords in the file. If salts are present, then the attacker would have to compute hash(salt[a] . attempt), where "." denotes concatenation, compare against entry A, then hash(salt[b] . attempt), compare against entry B, and so on. This defeats "reusing" hashes in attempts to crack multiple passwords.
Salts also combat the use of rainbow tables for cracking passwords. A rainbow table is a large list of pre-computed hashes for commonly-used passwords. For a password file without salts, an attacker can go through each entry and look up the hashed password in the rainbow table. If the look-up is considerably faster than the hash function (which it often is), this will considerably speed up cracking the file. However, if the password file is salted, then the rainbow table would have to contain "salt . password" pre-hashed. If the salt is long enough and sufficiently random, this is very unlikely. Unsalted passwords chosen by humans tend to be vulnerable to dictionary attacks since they have to be both short and meaningful enough to be memorized. Even a small dictionary (or its hashed equivalent, a rainbow table) has a significant chance of cracking the most commonly used passwords. Since salts do not have to be memorized by humans they can make the size of the rainbow table required for a successful attack prohibitively large without placing a burden on the users.
More technically, salts protect against rainbow tables as they, in effect, extend the length and potentially the complexity of the password. If the rainbow tables do not have passwords matching the length (e.g. an 8-byte password, and 2-byte salt, is effectively a 10-byte password) and complexity (non-alphanumeric salt increases the complexity of strictly alphanumeric passwords) of the salted password, then the password will not be found. If found, one will have to remove the salt from the password before it can be used.
Additional benefits 
The modern shadow password system, in which password hashes and other security data are stored in a non-public file, somewhat mitigates these concerns. However, they remain relevant in multi-server installations which use centralized password management systems to "push" passwords or password hashes to multiple systems. In such installations, the root account on each individual system may be treated as less "trusted" than the administrators of the centralized password system, so it remains worthwhile to ensure that the security of the password hashing algorithm, including the generation of unique "salt" values, is adequate.
Salts also make dictionary attacks and brute-force attacks for cracking large numbers of passwords much slower (but not in the case of cracking just one password). Without salts, an attacker who is cracking many passwords at the same time only needs to hash each password guess once, and compare it to all the hashes. However, with salts, each password will likely have a different salt; so each guess would have to be hashed separately for each salt, which is much slower since hashing is generally computationally expensive.
Another (lesser) benefit of a salt is as follows: two users might choose the same string as their password, or the same user might choose to use the same password on two machines. Without a salt, this password would be stored as the same hash string in the password file. This would disclose the fact that the two accounts have the same password, allowing anyone who knows one of the account's passwords to access the other account. By salting the passwords with two random characters, the odds are that even if two accounts use the same password, no one can discover this by reading password files.
See also 
- Password cracking
- Cryptographic nonce
- Initialization vector
- Padding (cryptography), sometimes confused with salting
- "Spice" in the Hasty Pudding cipher
- Rainbow tables
- "ISC Diary – Hashing Passwords". Dshield.org. Retrieved 2011-10-15.
- Morris, Robert; Thompson, Ken (1978-04-03). Password Security: A Case History. Murray Hill, NJ, USA: Bell Laboratories
- Wille, Christoph (2004-01-05). "Storing Passwords - done right!". Unknown parameter
- Primer on Unix password structure
- OWASP discussion of hashing java with salt
- OWASP Cryptographic Cheat Sheet
- How to effectively salt a password stored as a hash in a database