I have a set of single channel expression data from Nimblegen. On these arrays, there are no mismatch probes. We estimate background with a subset of probes that target a very different organism and are essentially negative controls. These 'background' probes are randomly distributed so they broadly sample background across the array, and generally provide some signal above just empty glass. Data are between-slide normalized in limma with vsn. I want to get a significance value for each gene just to say if it is expressed or not (above background), with a simple test. For each gene, for each experiment, I compare the probeset (20 probes per gene) to my negative controls (~ 2000 probes) as vectors by Mann-Whitney test: vector1 = intensities for probes targeting the gene vector2 = intensities for negative controls wilcox.test(vector1,vector2) If I have 5 replicate arrays, I can perform the test as above just by combining intensities across experiments and comparing experimental probes to controls as two vectors. I'm not sure this is really an appropriate way to handle replicates. Is there another function like wilcox.test, that can compare two matrices where replicates are in columns, and each matrix has a different number of rows? Thanks for your help! Dave Sturgill davidsturgill[AT]niddk.nih.gov