In the suggestion pot, Clem asks 
How about an article on ways to handle sense/detect type rolls? The spectrum goes from “player knows what attribute to roll against, what number is needed for success, and what is being checked for” to “player knows attribute only” to “just roll the dice” to “gm rolls openly” to “gm rolls secretly” to “gm rolls before the session and applies the result if/when needed”. Perhaps some discussion on which methods players are likely to accept and when to use them and whether it ever makes sense even with good roleplayers to say “you missed your Vision roll and you don’t see the python in the tree”.
Are Sensing Rolls Different?
In many instances, yes, a sensing roll is different from an attack roll or a climb check. By announcing that players need to roll a sense, you’re introducing the idea that there’s something to sense. Announcing that a player needs to roll for his PC to hit or to see how well they hide is usually a reaction to the PC’s action.
Despite that contrast, how you handle announcing target numbers is often constant across the game. (Everyone alters their general guidelines to achieve specific effects; here I’m talking about standard procedure.) If your players enjoy the numbers and metagame considerations, you’re likely to announce target numbers and identify threats; while if they’re concentrating on getting in their character’s head, you’re more likely to align player and PC information.
Keeping the Players in the Dark
Some players really enjoy having the same information as their characters and no more. While some things are quantified by any game system, these players often like getting into their character’s headspace and feel that extra knowledge is just an impediment– something else to work around.
For players like these, your “just roll the dice” suggestion tracks most closely if you like keeping the dice in the players hands. You might find it even more rewarding for the GM to roll stealth (or effective stealth, the DC of the trap/object concealment -10 in D&D) against the passive perceptions of the PCs. [In 3.x D&D, roll the stealth of the object against 10+the PC’s Spot (or equivalent) skill.] Since the GM often rolls for things throughout the night (random encounters, warming up the dice, etc.), it draws less attention to a specific moment or location than asking a player to roll. Other good methods include having the players provide a dozen rolls before the session starts– when it’s time for the player to roll a sense, just quietly cross off the top roll from their list and use that value. (You’ll want a copy of the PC skill bonuses so you can add them to the roll you scratch off.)
This approach is a polar opposite of trying to align player and character knowledge, as in the example above. The GM embraces metagame knowledge, trusting the players to use their out of character information to make the game more interesting. The GM clearly announces what’s being rolled: “OK Simon, your character Reyna needs to make a DC 17 listen check, or the muggers in the alley will surprise her.”
An advantage of this system is that the player can tweak the roleplay of the PC to match the metagame roll. So if Simon rolls a 6, with a resulting total of 13, he might say, “As Reyna cuts into the alley, she’s still thinking about the strange corpse that attacked her earlier that night. She’s lost in her own world, her feet taking her home almost on autopilot as she replays the scene in her mind.” An advantage is that you allow the player to define how their character succeeds or fails. This can let a player define their high search checks as consistent attention to detail (like Sherlock Holmes) or as bumbling that just happens to always illuminate the critical clue (like Shaggy and Scooby).
Unlike many compromises, picking a point between the two options above is usually worse than the extremes. Giving partial but incomplete information often strips away the advantages of both systems above– by announcing a roll, you disrupt the immersive player’s attempt to stay in their PC’s head, while hiding target numbers (or skills) doesn’t allow the PC to roleplay success or failure until they know the result. Further, by not announcing the effect of failure, the game hangs at that moment… which some players will take advantage of.
How many players have been in a group where the GM calls for spot checks and immediately the PCs are described as being on high alert at that very moment? Or when a thief examines a chest and rolls a 2 on the search skill, the thief responds, “Well, I don’t trust that so important a burgher wouldn’t have a trap on his chest…” and they roll again? [One solution to the “rolling again” problem is Burning Wheel’s Let it Ride— you roll a skill once and keep applying it to every opposed check until circumstances change.]
What will your players accept? In most cases, players expect rolls for their senses to match their other rolls in the game. If you typically announce the DC of an skill roll, you’ll probably want to explain why you’re treating senses differently. Many of the reasons will be apparent as soon as you bring them up– but not until then.
The hardest time to get buy in to restricted information is in the moment. If you wait until the rogue fails his search roll and deny him a chance to “examine the walls” instead of continuing his movement onto the trapped square, it may sound like you’re railroading him into damage. Or, worse, the game may stall while you discuss whether specific description elements are enough to justify a different action or provoke another roll.
In Your Game
How do you and your group handle sensing skills? Have you tried different styles and come on one that you like best? Did the post work for you Clem, or is there another angle you’d like to see discussed?
My thinking has changed a bit over time… three years ago I responded to Martin’s Blogging for GMs Project  with this post  about sensing and spotting. (Though I worried more about what the target numbers should be rather than technique in the post above.)