PRIME DIRECTIVE: Regarding some of the above posts, If you are having to strategically "leave gaps" or otherwise write bad documentation for the purposes of monkey wrenching your replacement or making yourself indispensable, you suck to the 10th power. I have dealt with this sort of fuckery more times than I can count over the years, and every time I clearly see the signs of a small mind at work. Don't be that guy.
I've have to do this routine a gazillion times in my role as an small/midsize biz consultant. Here's the formula as I see it:
Find the keys to the kingdom, and document these. Don't worry about getting to the nitty-gritty yet: just find the info that will let you find everything else. Be extra careful to track down any crypto-related stuff (keys, passphrases) that can't be replaced or cracked. The further the old IT person recedes into the past, the harder this crap is to track down. Identify the scariest bits of the network as quickly as you can after you get hired and trumpet to the hills about how fragile, dangerous, and not your fault they are :-) Document all of this in plain-old-textfiles or something stupid simple, with bonus points if you keep it in version control.
Set up some bug-tracking/ticketing software and use that to track all of your day-to-day documentation and troubleshooting. Redmine is my personal favorite, but RT and Trac are also good choices. They have simple, built-in wikis that are perfectly sufficient for this purpose. Use the time-tracking and project management features in the software: when the boss asks you where all of your time is going, you run a report and show him. Track every minute of your day: this time is excellent leverage for you when dealing with management.
Examine the backups of the system (or implement them, worse) from ground zero. Use the backup audit as the trail of breadcrumbs that your documentation follows.
Beg, borrow, or steal a chunk of hardware that you can stick Xen/HyperV/VmWare/AcmeHypervisor on and start test restoring various systems/apps/environments. Document the hell out of the test restore process. That's the most precious documentation that you can possibly have.
Don't document things that document themselves. You're much better off paying $299 for a copy of LanSweeper or the like to reach out across your networks and document all of the mundane details in real-time so that you can focus on making shit work right. I've seen a million cases where the IT guy spent a month making a beautiful set of Visio network maps that became useless a month after they were created. That's a waste of your otherwise precious time.
Now, start making recommendations about how to fix the fragile/scary/dangerous systems. Use your ticketing/project management app to track your recommendations and leave a paper trail of your process. There will be a fair amount of CYA involved here - you're going to recommend that the boss spend a bunch of money on $x, so you had better document the reasons for it with care.
Whenever you make a recommendation that the company buy product $x to resolve problem $y, document it very very clearly in your system, and if the bosses nix your recommendation, document that in writing too. Yes, more CYA, but as the IT guy you're often the staked goat when something goes wrong. You need to be able to PROVE that you had recommended a sane course of action.
Last: Try not to stress. IT is fun if you do it right.