Microarray analysis provides simultaneous measure of gene expression levels in biological experiments. One difficulty of microarray analysis is the uncertainty evoked by large amount of instability due to various kinds of systemic and random fluctuations that can degrade the analysis. This thesis develops a dependency representation on the spot of a microarray design that we called consigned correlation pattern. A pattern distance is applied to evaluate a set of microarray data that can also identify spot groupings. A prediction rule from the spot signals is designed to predict the class of a microarray that can identify irrelevant range of the spot signal values. The advantage of such an approach is that global interdependency information is incorporated to evaluate the stability of the spot signals in the microarray design. Experiments using a relatively large set of microarray samples with a small number of spots on genes from six food-born bacterial strains are performed. The results yield interesting gene inter-relationships indicated by the spot signals. The groupings of the spot signals also reflect the effectiveness of the designed probes in identifying the bacterial strains. The prediction rule is found to be less susceptible to data with a large amount of irrelevant signals.