There's a whole class of encounter that doesn't really involve combat, or even an adversarial interaction with an NPC. Instead, your PCs are tasked with piloting a boat down a stretch of rapids, or driving a carriage through a chase sequence, or working a series of Gather Information checks to track down the location or identity of some Plot Token that is needed to advance through some phase of an adventure. It could be anything, really; the only thing we're concerned with is that there's something happening that pits the PCs against something that isn't a combat encounter and isn't an NPC.
The DMG isn't terribly helpful, here. There's plenty of instruction if you need help figuring how to count multiple creatures with the same (or different) CR to derive an Encounter Level that will allow you to gauge how hard the encounter should be. And if you're dealing with a non-combative encounter against an NPC or monster, it's actually pretty explicit that you can still award XP based on this underlying mechanic--CR and EL are used to gauge the amount of treasure and XP that should be awarded if the PCs "overcome" the encounter, not necessarily what they get for looting corpses and strongboxes.
I think this leads a lot of DMs just to rely on gut and eyeball. But there's an alternative, which requires you to work backward from the rules for traps. So let's look at a trap, for just a second. The following is the very first exemplar from the SRD:
Basic Arrow Trap: CR 1; mechanical; proximity trigger; manual reset; Atk +10 ranged (1d6/×3, arrow); Search DC 20; Disable Device DC 20. Market Price: 2,000 gp.
We can disregard anything about this that doesn't have bearing on the trap's CR. So we can disregard that this is a mechanical trap, how it's reset, and how it's triggered. Those things are solely involved in pricing the trap, and they don't have anything to do with how difficult the trap is to overcome. Traps are rated for challenge based on how hard they are to detect (Search DC), how hard they are to disable (Disable Device DC), how reliably they take effect if triggered (Reflex Save DC OR attack bonus), and how damaging they are. Each of these categories gets a scale that determines how much it contributes to the trap's final CR. And better still, except for damage potential, they all use the same scale.Difficulty Class CR Modifier Less than 15 -1 16 to 24 +0 25 to 29 +1 30 and greater +2
If a trap uses attack rolls, then in essence it still uses the above figures; an attack bonus of +6 to +14 counts as +0 CR, and higher and lower bonuses modify CR in the same ways. You can convert the attack bonus to the DCs listed here by adding 10 to the attack bonus. Simple enough.
Damage scales in multiples of 7 points. In a rare exception to the general rules about rounding in D&D, this scale rounds upward when that would be appropriate for standard mathematics. In our example trap, the entirety of the CR 1 listed is derived from its damage output--the attack bonus, Search DC, and Disable Device DC all are in the +0 CR ranges.
CR 0 doesn't exist in the 3.5 rule set; either something presents a challenge, or it doesn't; very minor challenges merely present a lesser challenge than a full CR 1. So if we cared to do so, we could create a CR 1/2 trap from the example by eliminating its damage-dealing capability and leaving everything else the same. It'd work out fine, so long as there was still some negative consequence for falling victim to the trap--maybe it trips up its victims or sets off an alarm. And really, we could keep pushing the CR lower by reducing DCs. A non-damaging trap that is detectable with a DC 15 Search check, disarms with a DC 15 Disable Device check and which relies on a +5 attack bonus would have a CR of approximately 1/6.
Conversely, we also can use this framework to create a non-damaging trap that has a higher CR. With a +20 attack roll, Search DC 30, and Disable Device DC 30, the framework yields a CR 3 trap. In fact, we could extend this framework to encompass all sorts of skill challenges. As long as there are roughly three checks, and all of them are against roughly DC 16-24, the outcome is a CR 1/2 challenge. In fact, we could immediately adapt this principle to create a CR 1 skill challenge in the form of a tightrope crossing. To overcome the challenge, the PC must successfully negotiate three consecutive Balance checks against DC 20. If he fails a check, he immediately falls 20 feet, suffering 1d6 points of damage. If we want to make the tightrope only ten feet above ground but keep the rest of the difficulty, it deals 1d6 points of non-lethal damage instead, then we have CR 1/2 challenge.
It's also somewhat feasible for us to use this mechanism to control common activities like the use of Gather Information to unearth the whereabouts of a fugitive NPC. Let's imagine that your PC is interested in tracking down a skilled forger who can provide the appropriate fake credentials to allow him to pose as a military officer. We can lift the basic mechanics for this task almost directly from the Urban Tracking feat. Let's say that the search takes place in a "large city" of approximately 13,500 residents. This sets our base Gather Information DC to 15. We'll say that the primary race of the city is human, followed by dwarf, but our fence is a gnome. So we'll modify the DC downward a couple of points, because gnomes are rare enough to stick out in this town. And then we'll add 5 to the DC, because the gnome knows this perfectly well, and as a forger he does his best to maintain a low profile anyway. We get a final Gather Information DC of 18. That gets us a +0 CR modifier.
The PCs aren't going to take any damage if they bomb a Gather Information check, so right now, we're sitting at a CR 1/2 encounter. If we look at Urban Tracking, though, we see that finding someone in a large city takes 2d4 checks (average 5). So this probably works out to a gentle CR 1 challenge because of the overall number of checks required. The low CR is an outcome of the lackluster DCs involved in the Urban Tracking application of Gather Information, which is designed with the idea that it should be hard to fail these checks if you've invested significant skill ranks into Gather Info—and also because it's a "plot essential" bit of activity. If your adventure is written so that it depends on your success at urban tracking in order to move it along, then failure means you don't have an adventure after all. This would be a problem by the lights of most gaming groups, so the DC is artificially a bit low for this example.
But also, this is more generally an outcome of the limitations of the trap mechanics underlying this approach. A skill challenge against DC 34 is still only about CR 6, even though many characters would struggle even in the middle levels of the game. A better approach to make a skill challenge more difficult is by packing two challenges together. Returning to our non-lethal tightrope challenge, we can make it a CR 2 challenge by asking for one of the checks against DC 25 instead of DC 20, or perhaps we also intersperse a Tumble challenge (also relying on two checks against DC 20 and one against DC 25) with it, to obtain an EL 4 skill challenge made up of a pair of concurrent CR 2 challenges.
Despite the challenge involved in making this framework yield single-skill challenges of CR higher than 6, I think it has a lot going for it. Not least of all, it's very predictable and objective. You don't have to use your gut, or throw your hands into the air in frustration and conclude that you should just ignore XP and have the players gain levels as you decree it appropriate. Instead, it yields a very straightforward result. If you start with three checks per challenge, and say that you want the skill checks to have difficulty of X, Y and Z, then it spits out a CR that allows you to award XP according to established practices. It also allows you to combine skill challenges, either with one another or with a combat encounter (maybe the rogue is trying to get to a lever or switch, while the rest of the party is fighting a monster), and get a predictable combined EL as the output.