Although I will occasionally refer to them, I prefer not to use pre-published random monster tables. If I've decided that Orcs, for one reason or another, don't exist in my world, a load of 1st-3rd level charts, from the 1st Edition DMG to the Pathfinder Bestiary, suddenly have gaping holes in them. You can substitute brigands, say, when that happens, but at that point, you're just making things up anyhow, unless you pencil in a substitution for every occurrence of the word "orc" in the table.
Needless to say, such behavior is beneath decent, civilized folk like us. So maybe you just look up and down the table until you see something that strikes your fancy. Nothing wrong with that; we've all done it before.
Nonetheless, having a table you can roll some dice against keeps you honest, and also keeps you from returning to the same kind of encounter by force of habit. Also, by not being tied to a specific location, wandering monsters give the illusion that the game world goes on even when the PCs take the day off; those Bugbears obviously don't live in the Ghoul's Crypt, they're probably looking to loot it themseves. (That, or they've made a very bad choice of shelter coming in from the rain.)
The solution to the problem of inappropriate results is to come up with your own tables. And there are two parts to each basic table: the "stuff", and the "numbers".
The "stuff" is just that, whatever's going to show up after you've rolled on that table. A treasure hoard, perhaps. Or a monster. Or an NPC personality trait. This part is not particularly hard to come up with; you just start listing things that you want to see in your game. A table is nothing more than a list with some weighted probabilities attached to it. You list the things that are going to be interesting, not necessarily what is most likely to be there. An NPC may have a secret fear of snakes, but if it's not going to come up in the game, why waste a line on the table?
The trouble is with the numbers. You want something that you can easily roll on dice, and that reflect your (subjective, approximate) feeling of how likely it is to come up. (Note how I said, just a paragraph before, not to list things by how likely they are to come up; that's true for the list part of creating a table, but not for the numbers part.) There are countless ways of doing this, but for the beleaguered DM, it boils down to two old friends: the Bell Curve, and the Linear Distribution.
If you're fine with Linear Distributions, then you're good if your list contains 4, 6, 8, 10, 12, 20, or 100 objects. Just roll the requisite die. Most of us know you can also simulate, say, a d16 by rolling 1d8 and a 1d6; if the d6 comes up 4-6, add 8 to the d8, otherwise just read it as it is. This process can be generalized to get linear distributions of 16, 24, 30, 36, or 1200 results, but I'll leave those as exercises for those so inspired.
But what about Bell Curves? Are we stuck with mimicking D&D stat rolls? We all know that 3d6 create a Bell Curve, and that's great if you have sixteen possibilities that you want to select from; but if you have twenty, it seems sad to relegate your score of encounters to a Linear Distribution just because you have a d20.
The trick is to realize that there are more ways to generate a Bell-Curve-like distribution than just 3d6; rolling 2d8+1d6-2 gives you a slightly distended bell curve in the range of 1-20. Rolling 3d4-2 gives you a curve between 1 and 10. 2d4+1d6-2 gives you 1-12. (There's no reason to subtract that two, if you don't want to; I just like to start from 1 for purposes of illustration.)
If you don't mind doing the math in your head, d3+d4+d6-2 gives you a curve from 1 to 11. The advantage of using an odd-numbered die (either by halving a d6 or a d10) is that the middle value will be uniquely more common than the rest. With regular polyhedrals, the middle two values will always be equally likely. The disadvantage is, unless you have a random number generator handy, you will need to do a slight bit more arithmetic in your head.
Three dice that are close to each other in terms of number of sides (like a d4 and a d6) work best when you want the probabilities fairly close to each other, but there's no reason you can't do, say, 6d4-5 to get a curve where the middle value (ten) is really more likely than the next two. It's probably best in most cases, though, not to obsess with how much more likely one value is to appear, and just realize that values closer to the middle will be more common than those farther away.
So now you have, say, twenty items on your list. And you have a table with lines numbered from 1 to 20. You assign the two most likely (or most/least desirable, it's your call) encounters to slot #s 10 and 11; you take the next two and put them in slot #s 9 and 12; and so on, until all 20 slots are filled. Optionally, you can put the more desirable encounters on the lower half (10 instead of 11, 9 instead of 12, etc.); then a successful recon strategy or a good Search/Spot/Survival roll can adjust the die roll down a point or two, rewarding the cautious player.
When you need to pull out the chart, roll 2d8+1d6-2 and check that number against the table. (I recommend maybe noting the required dice and math at the top of the sheet, just to make it easy to remember.) When you roll a 20, your players will know you're honest, and aren't just siccing Orcus on them because they handily dispatched your previous encounter.