You might be able to get away with using regular expressions, but a more robust solution would involve using HTML::Parser (alt.) or the kinder, gentler HTML::TokeParser (alt.).

Update: Searching for either module in Super Search should turn up lots of useful nodes, but this tutorial by crazyinsomniac might be a good place to start.

Update^2: I'm surprised Ovid didn't mention it himself, but crazyinsomniac kindly pointed out HTML::TokeParser::Easy ... like H::TP wasn't easy enough. ;^)

    --k.