It's easy to create a mess in blender, either on your models wireframe, or the overall scene. I know about cases by experience, and some people found me on IRC asking about them lately, so I thought that doing an article about basic topology would help newbies out.

The things I am going to teach you now basically apply to low-poly modelling. I don't think that subsurf will do magic and solve your issues, mistakes, and that your starting mistakes won't show! A good start is important, setting up the base as perfectly as you can, is the most important thing to do. If you have a badly shaped starting mesh, or you have too many objects lying around, you wont be able to see the system trough and you will feel lost. You wont be able to track your materials, textures neither, or anything, and you will become frustrated quickly and might as well give up on the job. A good start is important, and its also important that you dont have any innecessary edges, loops, faces or vertexes on a hard surface.

Three cubes, all of them look like the same, but two of them is eating your CPU innecessarily. The icospheres floating above them might show why that could happen. If we switch to edit mode, you will see what is going on:

You can't see the difference in the cubes, and its minimal on the icospheres as well. The icosphere to the right has lots of subdiv levels, and the icosphere to the left, with its simple subdiv and smooth shader almost looks like the same. But if we do a render, we can see how minimal is the difference, and notice that the subdivs on the cubes don't have any effect at all.

If we use the subsurf modifier, we have calculate how many round, organic shapes are we going to use, and we have to clean the hard surfaces later on. And we should calculate the distance to the camera as well, because if we took the left icosphere as an example, believe it or not, from a distance it is the same as the right one and it doesn't even renders that long.

But, since I talk about those balls a lot, lets take a closer look. The scene below has icospheres only. The middle one is a simple icosphere, the one to the left has a simple subdiv, and the right one has 3 levels of subdiv. The difference is minor, even in the preview window:

In edit mode:

 

You can see the difference here. We have lots of triangles, and we dont need the half of them on the top right ball. Lets render the scene:

You can see the difference: middle one has flat, cubical reflection. If we use 1-2 levels of subdiv, it looks quite correct. But if we use more than enough, it will reflect the world in a strange way, and that is only because it has too many triangles. To summarize this part: if something is not close to the camera, its innecessary to do too much geometry. This applies to a close object which is not lit that well, because textures can do the trick. The most important is that do not leave anything on flat surfaces, because that's just a waste of rendertime. Here is an image, which shows that how much detail can be done with low-poly modeling techniques:

And now, lets take a look at using objects and groups. If we start Blender, we have three objects: a lamp, a camera and the cube. If we add more elements to the cube in edit mode, that will still count as a single object. If we duplicate the cube in edit mode, we end up with a single object with two meshes. If we have a sphere and a cube on the scene, we have sphere object with a sphere mesh and its data, and we have a cube with a cube mesh and its data. If the sphere object has a red and a green sphere in it, we end up with 3 meshes but we still have only 2 objects.

If we want to use objects and groups logically, we just need to take a look at our everyday life: a chair shouldnt have all its parts set to single objects, we can have it in a single object with its feet, arms and other parts. And if we have an array of chairs, that could be a single object as well. And if we have a table set for the chairs we can group that. It all depends on your goals, and your asset management pipeline.

The images below are showing you one of my models arrangements. I just mirrored the bones, and thats an object. The main rings are a single object as well, the gemstone too and the stairs are too. Thats because we can now use the stairs, the rings, the bones or even the gemstone for other models or scenes without having to make separate files for them. But if we wont do that, it can be merged into a single object anyway.