Image Analysis
In general, our analysis methods for assessing differential expression are based on quality control and replication. Each array has its own internal level of quality and stability. We assess this on an individual basis by looking at replicate spottings of the same gene at different locations on the array. Disagreements between replicates are due to random variation (which we're not interested in) and give us some idea of the scale a difference must achieve before it is "big enough" to be believed. This scale of variability is linked to the overall intensity of the signal; low-intensity spots are measured with less precision.
Dirtier arrays have more disagreement between replicates, and hence have to have larger differences to be believed. In addition, there are trends in the efficiency of hybridization which can be estimated and removed using replicates. Checking and calibration is done both with each dye channel and with the overall ratios.
Types of samples have similar associated levels of variability, and these can be estimated using replicate arrays in much the same fashion.