It looks like the superseries to which this CEL file belongs on GEO (see here) contains both mouse and rat arrays. Is it possible you're intermingling them? Affy would probably peg the CDF geometry off of the first CEL it happens to read.

If that's not it, check the first 20 lines of a CEL it will read (you can just use more on the CEL file and look at the header) and compare that to the problematic CEL file. The header information should look similar though not identical to a working one.

If you're using ReadAffy to load your cell files, then you can pass a list of CEL files to read rather than sucking up all files in the working directory. Alternately, you can rename a file "foo.CEL.doesNOTwork", but it would be better to figure out what the real problem is.

Yup, thanks (even if I prefer less :) ). They were all mouse data, I'm already passing a list of cel files to ReadAffy and not a single directory: I would like something like a try/catch around single cel loading to solve this problem once and for all (the R script which normalizes should be able to detect the ``wrong'' cels, print out a warning and go on with the good ones) without manual intervention - I will read a little bit more of the affy documentation to see if it's possible. Let's see these headers...

...and here it is, my fault (what a surprise...): it appears that the CELXXX_RAW.tar for GSE14499 even if they are not in the filelist and are in a different GSE also has rat data and I didn't double check its contents. Thank you for your help, next time I won't trust .tar from a single GSE :)