Correlated subqueries are a useful aspect of SQL that didn't come naturally to me. noncorrelated subqueries are not dependent on the containing query, and are executed only once, before the containing query. In contrast, correlated are dependent on the containing query, and are executed once for each row of the containing query.
Here's an example, taken from the book Learning SQL.
SELECT c.cust_id, c.cust_type_cd, c.city FROM customer c WHERE 2 = (SELECT COUNT(*) FROM account a WHERE a.cust_id = c.cust_id);
This returns info about customers who have exactly two accounts. For each customer, the subquery runs through every account, executes a filter to only include accounts associated with the customer in question, and returns the number of accounts meeting this condition. This result is then compared to the number 2 in the containing query. If it is 2, then that customer is included in the result of the containing query.