People will usually reply to my criticisms about SQL NULL by pointing out that it is logically consistent, which is true, but it's not the only logically consistent NULL we can define. I know the theory behind the SQL NULL quite well, well enough to observe just how rarely it matches SQL practice.
In Date's book "Database in Depth" he gives some wonderful examples of how NULL in SQL breaks things. One I recall involved a couple simple tables and a simple question about the data. There were something like half a dozen or so reasonable ways one might write a query to answer the question--and there were many different answers depending on the exact query.
"Database in Depth" has been updated and refocused more on SQL, and is now called "SQL and Relational Theory". He sums up in the latter book the problem with NULL thusly:
To sum up: If you have any nulls in your database, then
you’re getting wrong answers to certain of your queries.
What’s more, you have no way of knowing, in general,
just which queries you’re getting wrong answers to and
which not; all results become suspect. You can never
trust the answers you get from a database with nulls.
In my opinion, this state of affairs is a complete
showstopper.