You normalize your data internally in microarray experiments. I'd suggest nomalizing your data to the 50th percentile in each chip. If you work with affymetrix data, there is already at least 11 probes present for each probesets so no need for additional normalization.
Also, if you're planning to use 18S, good choice but a good idea is to check its expression profile in your microarray. Try to find the one gene with less differences between samples and use it as a normalizer. It doesn't have to be one of the common housekeeping gene as long as it is constant.
