Thanks to all who have sent comments and suggestions. I would like to ask one more question. I am doing the filtering to prepare a dataset for some several class discovery methods that blow up with relatively large datasets. I have about 60 tumor samples coming from 7 known biological classes, but comparing the means of these groups to look at fold change would bias the method. Assuming intersample normalization is it justified to compare the overall highest with the overall lowest as a fold change measure or is there a better way of doing this? Thanks Peter