Difference between USING and ON in Oracle SQL

Not much. Of course, ON allows this, which USING does not:

SELECT a.col1, b.col2  FROM table a JOIN table2 b    ON (a.col3 = b.col4)                      ^

This is the ANSI join syntax, it is not unique to Oracle.

The difference for me is that you can paint yourself into a corner with the USING clause:

CREATE TABLE roster (mgrid INTEGER, empid INTEGER);CREATE TABLE emp (empid INTEGER, NAME VARCHAR2(20));INSERT INTO roster VALUES (1,10);INSERT INTO roster VALUES (1,11);INSERT INTO roster VALUES (1,12);INSERT INTO roster VALUES (2,20);INSERT INTO roster VALUES (2,21);INSERT INTO emp VALUES (10, 'John');INSERT INTO emp VALUES (11, 'Steve');INSERT INTO emp VALUES (12, 'Mary');INSERT INTO emp VALUES (20, 'Ann');INSERT INTO emp VALUES (21, 'George');INSERT INTO emp VALUES (1, 'Pete');INSERT INTO emp VALUES (2, 'Sally');SELECT r.mgrid, e2.name, e1.empid, e1.name  FROM roster r JOIN emp e1 USING(empid)                JOIN emp e2 ON r.mgrid = e2.empid; 

In the above select, you get an ora-25154, "column part of USING clause cannot have a qualifier".

If you remove the e1.empid qualifier, as in:

SELECT r.mgrid, e2.name, empid, e1.name  FROM roster r JOIN emp e1 USING(empid)                JOIN emp e2 ON r.mgrid = e2.empid; 

You get an ORA-00918 error, "column ambiguously defined".

You have to use:

SELECT r.mgrid, e2.name, e1.empid, e1.name  FROM roster r JOIN emp e1 ON r.empid = e1.empid                JOIN emp e2 ON r.mgrid = e2.empid; 

The example is contrived, but when I was first exploring the join syntax I ran into this exact problem in a real situation. I have avoided the USING clause ever since. There is no advantage with the USING clause other than a few keystrokes.

ON includes all the columns that are being joined from the joining tables, so that you have two or more of the same column types in the resulting dataset.

But USING removes duplicates of the columns that are being joined to leave one of that type,which makes it similar to a natural join (except they don't have to have the same name) as highlighted in this question on natural joins

USING can be useful ;) for doing inserts with a subquery that joins tables (in my case flattening tables for a data warehouse star schema, where I only wanted one of the id columns).