I'm guessing efficiency isn't too much of an issue, so I'd split it up in parts:
a) Generate all Fib numbers up to 100.
b) Generate all Primes up to 100.
c) For each element of Fib, run through Primes and check if there's a match. If not, print.
Once you get it working, you can worry about optimization or combining parts to cut away some double work. Especially Prime generation can be done much more efficiently than a naive implementation. It's not going to matter much for "up to 100", but it's interesting stuff!