There are many great sources for this answer, you may want to try http://www.swcp.com/~jgentry/topo/unit3.htm or http://www.learntosubnet.com if that doesn't help please repost.

In a nutshell I think the most common practice would be to classify devices..

For example..

You setup 255.255.0.0

Now in your DHCP range you start the first ip at 50.1

Now you can say..

All Printers will be 10.1 - 10.254

All Hubs 11.1

All Switches 12.1

Etc Etc..

There are MANNNNY other reasons.. But I was just giving a brief example.

This is a good answer!

It is VERY important to note that if you set your subnet mask at 255.255.0.0 and you assign things to be in addresses like:

X.X.1.1-254 = servers
X.X.2.1-254 = pc's
X.X.3.1-254 = printers

All of these devices are actually in a single network as defined by the 255.255.0.0 subnet mask and you have not actually subnetted, rather you have administratively defined a standard numbering scheme.

This only actually becomes subnetting once you take your network and break into down into smaller networks, sub networks.

example:
-your network for your HQ is 172.16.x.x (16 bit network)
-you assign 172.16.1.x with a mask of 255.255.255.0 to DHCP
-you assign 172.16.2.x with a mask of 255.255.255.0 to dedicated pc's and servers
* you will need a router or dual homed device to get information from one subnet to the other and back again. The joys of subnetting

Essentially you have taken 1 network with a possible 16,000 addresses and you have provided for 254 networks (technically sub networks) with 254 devices each within the same address space.