We needed to return a random subset of rows from a database, and someone pointed out the newid() function.

SELECT *
FROM Company
ORDER BY newid()

For the application concerned it works well, but as it says in Random Sampling in T-SQL "However, it's evaluated by the operating system, not T-SQL. In addition, while unique identifiers can be ordered, the comparison functions aren't documented. Unlike Rand(), NewID() has not been designed for use as a statistically valid random number generator." so it obviously must be used with caution.