Here are some scripts to process Affymetrix .CEL files.
They run on a linux machine and need Perl and R
(http://cran.r-project.org) installed.

DISCLAIMER: THESE SCRIPTS WERE WRITTEN BY TOTALLY
NON-PROFESSIONAL PROGRAMERS, AND ARE VERY USER-UNFRIENDLY AND QUITE 
SLOW. THE AUTHORS ARE IN NO WAY RESPONSIBLE FOR THEIR USAGE AND
WILL NOT PROVIDE EXTENDED SUPPORT FOR THEM.

SCRIPTS:

(i) cdf2psc: converts a .cdf file into a .psc file.
    Needs to be run only once for each chip series.

Usage: cdf2psc OPTIONS

where OPTIONS are:
                  -cdf   .CDF FILE
                  -out   output (.PSC FILE)
example: 
cdf2psc -cdf MG_U74A_V2.CDF -out MG_U74A_V2.psc

(ii) cel2raw: converts .CEL files into .RAW files.

Usage: cel2raw OPTIONS

where OPTIONS are:
                  -cel   .CEL FILE
                  -out   output file (.RAW FILE)
                  -psc   .PSC FILE
		  
example:
cel2raw -cel file.CEL -out file.RAW -psc MG_U74A_V2.psc

(iii) raw2pcel: estimates background and so on.
     (actually (ii) and (iii) should really be combined
      into one step...)

Usage: raw2pcel OPTIONS

where OPTIONS are:
                  -i   RAW DATA EXTRACTED FROM CDF AND CELL FILES 
(.RAW extension)
                  -o   PROCESSED CEL FILE (.PCEL extension)
		  -e   epsilon for the background PM-MM < epsilon

example:
raw2pcel -i file.RAW -o file.PCEL -e 50 (50 is usually OK with the
                                         new scanner settings)

(iv) cel2ratios: converts two .PCEL files into ratios, plus other 
stuff.

Output file must have .RAT extension


Usage: cel2ratios OPTIONS

where OPTIONS are:
	-f1     first  .PCEL file
	-f2     second .PCEL file
	-o      output .RAT file
	-noout  do not output ratio file, only output the .REP file
	-p      compute Wilcoxon p-values for ratios
	-adj    adjust (normalization), 1 for global, 2 for local 
(robust but slow cf. $Nloess)
	-sgf    computes the log-ratio divided by the local variance 
(robust but slow $Nloess)
	-mm     default is PM only; 1 for PM-MM; 2 for MM only; 3 for 
all PM,MMs  
	-max    default=24, maximum number of probepairs per probeset 
(some probesets have more, i.e. in HG_U95)
	-nc     default is 3, noise cutoff, probes below nc*sigma_eff 
are discarded
        -qu     default is .5, quantile to use in LTS regression

Examples:
cel2ratios -f1 file1.PCEL -f2 file2.PCEL -o out.RAT -p -sgf -adj 1 -
mm 1 (uses PM-MM)
cel2ratios -f1 file1.PCEL -f2 file2.PCEL -o out.RAT -p -sgf -adj 1 
(uses PM only)

(v) several scripts produce .REP files with some additional info

(vi) TIP: use the foreach command, and so on, whenever possible.