One often has to submit a document (e.g., a paper or a dissertation) under some sort of constraint about its size. Sensible people set a constraint in terms of numbers of pages, but there are some that persist in limiting the numbers of words you type.
A simple solution to the requirement can be achieved following a simple observation: the powers that be are unlikely to count all the words of a document submitted to them. Therefore, a statistical method can be employed: find how many words there are on a full page; find how many full pages there are in the document (allowing for displays of various sorts, this number will probably not be an integer); multiply the two. However, if the document to be submitted is to determine the success of the rest of one's life, it takes a brave person to thumb their nose at authority quite so comprehensively...
The simplest method is to strip out the (La)TeX markup, and to count what's left. On a Unix-like system, this may be done using detex and the built-in wc:
detex <filename> | wc -wThe latexcount script does the same sort of job, in one "step"; being a perl script, it is in principle rather easily configured (see documentation inside the script). Winedt (see editors and shells) provides this functionality direct in the Windows environment.
Simply stripping (La)TeX markup isn't entirely reliable, however: that markup itself may contribute typeset words, and this could be a problem. The wordcount package contains a Bourne shell (i.e., typically Unix) script for running a LaTeX file with a special piece of supporting TeX code, and then counting word indications in the log file. This is probably as accurate automatic counting as you can get.
This question on the Web: http://www.tex.ac.uk/cgi-bin/texfaq2html?label=wordcount