Surprise!

Surprise is a bit of a funny rule -- One wonders exactly what occurred to Gygax/Arneson to require the 2-in-6 chance for any side to be surprised as combat commences. For example, is it reasonable for an adventuring party, armed to the teeth and invading a particular location, to be surprised one-third of the time they smash down a particular door that an enemy is within? I suppose (a) it's at least balanced on both sides, and (b) it's more legitimate in the context of a mostly-empty dungeon as per OD&D.

Here's an interesting and significant difference between the OD&D and AD&D rules for surprise. In each case, the chance is a base 2-in-6 per party, and short-circuited by warning signals such as light, noise, and (notably) ESP. The effect of surprise differs, however. In OD&D:

Surprise gives the advantage of a free move segment, whether to flee, cast a spell, or engage in combat. If monsters gain surprise they will either close the distance between themselves and the character(s) (unless they are intelligent and their prey is obviously too strong to attack) or attack. For example, a Wyvern surprises a party of four characters when they round a corner into a large open area. It attacks as it is within striking distance as indicated by the surprise distance determination which was a 2, indicating distance between them was but 10 feet... The Wyvern may attack once again before the adventurers strike back. [OD&D Vol-3, p. 9-20]

So the effect here is to give one ("a free move segment") advance round of action. In addition, the monster gets a second attack routine ("may attack once again"). This latter clause requires some interpretation; personally, I interpret it as giving automatic first initiative in the subsequent normal round of combat. This makes a lot of sense, otherwise surprise basically degenerates to the same as simple first initiative, anyway. Here's the rather different effect in AD&D:

Each 1 of surprise equals 1 segment (six seconds) of time lost to the surprised party, and during the lost time the surprising party can freely act to escape or attack or whatever. If both parties are surprised, then the effect is negated or reduced... Example: Party A is surprised only on a roll of 1, but party B surprises on 5 in 6 (d6, 1-5) due to its nature or the particular set of circumstances which the DM has noted are applicable to this encounter... Assume A rolls a 4, so it is surprised for 4 segments... [AD&D DMG, p. 61-62]

So here, the primary alteration is the allowance for surprise chances to vary, possibly as high as 5-in-6 or more (see: Spider, Huge), and for each "pip" of the surprise die to indicate an additional free round for the attacker. As the example above indicates, this could result in as many as 4 or 5 free rounds of attack from certain kinds of monsters, without any return action from the PCs! This always seemed overwhelmingly lethal to me, and it's one thing I could never broach applying in actual play (sometimes fudging dice when I was in my younger, OCD ur-text mode). An additional subtlety pops up in the example of melee:

As party B is surprised for 2 segments, party A has a chance to hit in each segment as if they were full rounds (this does not apply to spell use, of course)... [This resolved:] Now initiative dice are rolled, and party A's score is lower, so party B gets to react to the assault. [AD&D DMG p. 71]

As opposed to OD&D, the surprising party does not get automatic initiative after the surprise is over; they must dice for it, and either party might take the next action at that point. In play this seems clunky to me, as it puts the brakes on the action/pacing, and forces me to think about the effect of who goes next. It also has greater variation, since now a surprising party might get as many 6 unanswered actions in its favor.

This is also related to the difference in initiative systems between OD&D and AD&D: (a) OD&D refers back to Chainmail with its roll-initiative-once, then cycle-between-parties approach, while (b) AD&D rolls dice at the start of each and every round to see which party goes first, thereby creating lots of back-to-back doubled actions by certain parties.

Considering the marked distinctions between the games, I find that my preference is (no surprise!) for the OD&D rule, which is - as usual - more straightforward, easier to remember and apply, and has less fiddly game-breaking variation available to it. Similarly, I use the sensibility of the OD&D/Chainmail/Swords & Spells initiative system where the action simply cycles back and forth between parties (in my case, in order around the table) after the initial determination.