The bindings between key sequences and command functions are recorded in data structures called keymaps. Emacs has many of these, each used on particular occasions.
Recall that a key sequence (key, for short) is a sequence of input events that have a meaning as a unit. Input events include characters, function keys and mouse buttons--all the inputs that you can send to the computer with your terminal. A key sequence gets its meaning from its binding, which says what command it runs. The function of keymaps is to record these bindings.
The global keymap is the most important keymap because it is always in effect. The global keymap defines keys for Fundamental mode; most of these definitions are common to most or all major modes. Each major or minor mode can have its own keymap which overrides the global definitions of some keys.
For example, a self-inserting character such as g is
self-inserting because the global keymap binds it to the command
self-insert-command. The standard Emacs editing characters such
as C-a also get their standard meanings from the global keymap.
Commands to rebind keys, such as M-x global-set-key, actually work
by storing the new binding in the proper place in the global map.
See section Changing Key Bindings Interactively.
Meta characters work differently; Emacs translates each Meta character into a pair of characters starting with ESC. When you type the character M-a in a key sequence, Emacs replaces it with ESC a. A meta key comes in as a single input event, but becomes two events for purposes of key bindings. The reason for this is historical, and we might change it someday.
Most modern keyboards have function keys as well as character keys. Function keys send input events just as character keys do, and keymaps can have bindings for them.
On many terminals, typing a function key actually sends the computer a sequence of characters; the precise details of the sequence depends on which function key and on the model of terminal you are using. (Often the sequence starts with ESC [.) If Emacs understands your terminal type properly, it recognizes the character sequences forming function keys wherever they occur in a key sequence (not just at the beginning). Thus, for most purposes, you can pretend the function keys reach Emacs directly and ignore their encoding as character sequences.
Mouse buttons also produce input events. These events come with other data--the window and position where you pressed or released the button, and a time stamp. But only the choice of button matters for key bindings; the other data matters only if a command looks at it. (Commands designed for mouse invocation usually do look at the other data.)
A keymap records definitions for single events. Interpreting a key sequence of multiple events involves a chain of keymaps. The first keymap gives a definition for the first event; this definition is another keymap, which is used to look up the second event in the sequence, and so on.
Key sequences can mix function keys and characters. For example, C-x SELECT makes sense. If you make SELECT a prefix key, then SELECT C-n makes sense. You can even mix mouse events with keyboard events, but we recommend against it, because such sequences are inconvenient to type in.