Something that has been on my mind a lot has been random encounter tables. Each dungeon of my Megadungeon campaign has a different encounter table. And each has different mechanics. This bothers me.
First off, let's go through how random encounters are supposed to work by the book. I'm using the Rules Cyclopedia or BECMI here, but Basic D&D should all work the same when it comes to being in a dungeon. Every two turns, or every time the PCs do something that might draw attention to themselves, roll a d6. If it's a 1 (or a 6, if you want, the point is there's a 1-in-6 chance), then a random encounter is triggered. There's more rules for different situations, but that's the standard.
Then, you roll on the random encounter chart to see what you get. In the Rules Cyclopedia, for example, each dungeon level has a 1d20 table.
In a very good post on his blog Papers & Pencils, Nick LS Whelan reveals that he uses a 2d6 table. At the extremes—2 and 12—he puts Dragons and Wizards, respectively. Because they are awesome, dangerous, and should be an ever-present threat in a D&D game. What Nick introduces here is the weighting effect of the bell curve. Some monsters are more rare than others.
(I've always disliked the term "bell curve" to apply to anything number less than three dice, because you don't properly see the shape of the underlying Gaussian distribution. But anyway...)
Also, let Nick's velvety vocals whisk you away in podcast form here at Blogs on Tape. Complete with the theme to Dragon Warrior!
You could also easily convert the underlying bell curve into a percentile die roll. E.g., roll 1d100. A 1 is this monster, while a 2-3 is this other monster, etc. More on this below.
In Rappan Athuk (Frog God Games), the random encounter roll is combined with the result roll. As an example, on Dungeon Level 1, a single d20 is rolled and only results 1-4 have encounter entries. Therefore, the chance (1-in-5) is higher than your standard 1-in-6, but if that result is triggered, there's no need to roll another die.
On the one hand, this approach saves time by not having to roll multiple times. If you have very perceptive players, they may also get wise as to when you make one roll (your typical no-encounter) or when you make two (the encounter roll and the result roll). In this case, the one-roll method is advantageous in case you'd like to delay the reveal of the encounter. For example, maybe a result is rolled, but the creature is stealthy and would like to track the PCs for a while.
As a negative, you either get very few actual encounter entries in the table, or you need a much higher granularity.
Maze of the Blue Medusa has one of the best encounter tables I've seen in a product. It is a single d100 table, broken into a few columns that depend on where the party is in the dungeon. However, this usually just colors the frequency of individual encounters. Results from 1-20 are constant. 21-50 are encounter results that are location-dependent, usually referring to results 1-20. 51-74 is usually some sort of resource management, which is brilliant. For example, the PCs are hungry, or a torch goes out. Only on a result of 75-100 (slightly better than 1-in-4) does nothing happen at all.
This last table has me thinking about a few things.
First, there are a lot more potential encounters in this table than using any other method. However, I've found that most of the encounters in the Maze are more like vehicles for interesting action than straightforward combats. 2d4 screaming art critics is probably going to play out much differently than 2d4 kobolds.
Second, interesting events can be considered encounters, such as when that torch goes out. An encounter is really just a complication in the situation. If the torch goes out, then that's a surprise situation that you've got to deal with. Same as if you're messing with a door and a dragon saunters around the corner.
Lastly, the dungeon in question should inform the encounter table. Maze of the Blue Medusa has a dense encounter table because it is a dense dungeon. There's a lot of stuff there for the players to discover and interact with. Rappan Athuk is a sparse but deadly dungeon. Encounters there should be more rare, but they should usually scare the crap out of you.
So I am thinking about reworking my encounter tables to fit a unified paradigm. I'm leaning towards something like 3d6, where one of the dice is the "control" die, and the other two constitute a 2d6 table modeled after Nick's post above. Really, this is no different than the standard D&D method, except that the rolls are all made at once.
Perhaps later I'll go into other aspects I don't like about some random encounter tables, once I've had a chance to implement this idea into a game. But a lot of it might have its beginnings in my older post, Lessons from the Table.