Mew has charset guess mechanisms for both singlepart and multipart.
<Singlepart>
When you type `C-cC-m' on Draft mode, Mew guesses a charset from its body. On Bilingual Emacs, US-ASCII is chosen for 7bit charset while ISO-8859-1 is selected for 8bit charset. On Mule, a charset is chosen based on its internal representation.
<Multipart>
When you attach textfile on attachments region, the string "(guess)" is displayed as follows:
----------------------------- attachments -----------------------------
Multipart/Mixed 1/
1 Text/Plain(guess) CoverPage*
2 Text/Plain(guess) textfile
3 .
--------0-1-2-3-4-5-6-7-8-9--------------------------------------------
The difference of charset guess for singlepart and multipart is where the data locates. Singlepart is stored in buffer while each file of multipart is stored in disk. To guess a charset, Mew thus reads a file into buffer and uses the charset guess algorithm of singlepart.
On Bilingual Emacs, Mew reads a file as it is. So, if the file is 7bit, US-ASCII is chosen. Otherwise ISO-8859-1 is selected.
On Mule, Mew reads a file according to the local convention(i.e. auto conversion). The convention is really site dependent. In Japan, ISO-2022-JP, EUC-Japan, and Shift_JIS is neatly guessed and stored in buffer as internal representation for Japanese. Mew decides a charset from the internal representation. That is, you can safely attach ISO-2022-JP, EUC-Japan, and Shift_JIS text files. Furcations to decide local convention are called set[up]-<language>-environment. For more information about local convention, read their descriptions.
If you want to explicitly specify charset for a file, use `C' on attachments. A typical example is to attach ISO-8859-1 file in Japanese environment. The attachments region becames as follows:
----------------------------- attachments -----------------------------
Multipart/Mixed 1/
1 Text/Plain(guess) CoverPage*
2 Text/Plain(iso-8859-1) textfile
3 .
--------0-1-2-3-4-5-6-7-8-9--------------------------------------------
Note that `C' is not available on Bilingual Emacs.