RPKM and expression comparison

[kenosaki]

When you use RPKM for expression value and want to compare the value of multiple genes, what are the things you should be careful with?

Theoretically, RPKM is normalized expression value which used by the length of transcripts and the amount of reads per sample, so it looks OK for me to use this value for expression comparison between the genes with long transcript and the genes with short transcript.
But could be the nature is not so simple.

If you know some information regarding RPKM and expression comparison, please tell me.

Thanks

[jwfoley]

FPKM doesn't behave nicely and basically isn't directly suitable for statistical comparisons. Cuffdiff has a sophisticated solution: http://cufflinks.cbcb.umd.edu/howitworks.html#hdif