I had a situation where I needed to update or insert on a table according to two fields (both foreign keys) on which I couldn't set a UNIQUE constraint (so INSERT ... ON DUPLICATE KEY UPDATE won't work). Here's what I ended up using:
replace into last_recogs (id, hasher_id, hash_id, last_recog)
select l.* from
(select id, hasher_id, hash_id, [new_value] from last_recogs
where hasher_id in (select id from hashers where name=[hasher_name])
and hash_id in (select id from hashes where name=[hash_name])
union
select 0, m.id, h.id, [new_value]
from hashers m cross join hashes h
where m.name=[hasher_name]
and h.name=[hash_name]) l
limit 1;
This example is cribbed from one of my databases, with the input parameters (two names and a number) replaced with [hasher_name], [hash_name], and [new_value]. The nested SELECT...LIMIT 1 pulls the first of either the existing record or a new record (last_recogs.id is an autoincrement primary key) and uses that as the value input into the REPLACE INTO.