NLS_LANG is used to tell the database which characterset is used by the client so that the characters can be translated correctly when they are transmitted between client and database. There is an Oracle Metalink article that explains that explains the NLS_LANG parameter an contains a lot of references to important other articles:
NLS_LANG Explained (How does Client-Server Character Conversion Work?) (Doc ID 158577.1)
It is neither used to tell the client which character set is used by the database because this is not necessary. The client knows the character set of the database because it is connected to it.
Nor is it used to tell the database which character set it is using because the database knows which character set it is using.
So it is wrong simply to set the NLS_LANG to the character set of the database. It is a good way to corrupt your character data in the database
The purpose of the NLS_LANG variable is to tell the Oracle client which character set you unix/windows/... system is using.
Sot what is the value you have to set your NLS_LANG variable? On a Linux system do the following:
The character set of the unix system is defined by LANG, LC_CTYPE and LC_ALL. LANG is overridden by LC_CTYPE which again is overridden by LC_ALL. So in this case the must be NLS_LANG=AMERICAN_AMERICA.WE8ISO8859P1. This setting is independent of the character set of the database.
So first you have to check your locale and set the NLS_LANG appropriately.
There are some of other things that can go wrong e.g the database has already corrupt data or the terminl is not configured correctly.
Two references that explains the unix locale settings.