Dear conductors, I solved the normalization problem on my own. I was mainly using limma package. The data are obtained from 4 home-made oligonucleotide microarrays. They consist of about 800 50-nt-long DNA probes, complementary to the genes known to be implicated in disease development. First I’ve checked the quality of the data by creating MAplot and intensity profiles of both dyes using plotDensities() function. In order to compare changes of the data during normalizations I thought it would be nice to have boxplot, so I made one on my raw data. Creating spottypes was also a good idea. My MAplot looks much better and it’s more clear. I was’t sure which type of normalization I should choose, so I prepared all possible combinations of normalization methods. I was observing the changes on the MAplots and boxplots. I decided also to perform background correction. My results suggest that the normexp. and half are the most suitable methods for my dataset. They work nicely with loess and robustspline normalizations. It was also necessary to use normalization between arrays and the promissing results were given by scale and aquantale methods. The most suitable normalization model for my dataset is: Normexp.+ Standard+ Aquantile** and it gave much better results than half + „robustspline”+ scale (after all those operations the banana shape still was present). I couldn’t use the printtiploess, because there is a problem with my microarray dimensions. Each chip consists of two subarrays, with different size, so I decided to replace printtiploess with a loess method. I will also try to find a solution to my sctatistical problem or...I just ask my local satistician:) Best, B. [[alternative HTML version deleted]]