Microarray technology is a commonly used tool in biomedical research for assessing global gene expression, surveying DNA sequence variations, and studying alternative gene splicing. Given the wide range of applications of this technology, comprehensive understanding of its underlying mechanisms is of importance. The focus of this work is on contributions from microarray probe properties (probe secondary structure: ΔGss, probe-target binding energy: ΔG, probe-target mismatch) to the signal intensity. The benefits of incorporating or ignoring these properties to the process of microarray probe design and selection, as well as to microarray data preprocessing and analysis, are reported. Four related studies are described in this thesis. In the first, probe secondary structure was found to account for up to 3% of all variation on Affymetrix microarrays. In the second, a dinucleotide affinity model was developed and found to enhance the detection of differentially expressed genes when implemented as a background correction procedure in GeneChip preprocessing algorithms. This model is consistent with physical models of binding affinity of the probe target pair, which depends on the nearest-neighbor stacking interactions in addition to base-pairing. In the remaining studies, the importance of incorporating biophysical factors in both the design and the analysis of microarrays has been investigated. First, the impact of incorporation of a complete model of the hybridization equilibrium was tested. The results suggest that the use of probe `percent bound', predicted by equilibrium models of hybridization, is a useful factor in predicting and assessing the behavior of long oligonucleotide probes. However, a universal probe-property-independent three-parameter Langmuir model has also been tested, and this simple model has been shown to be as, or more, effective as complex, computationally expensive models developed for microarray target concentration estimation. The simple, platform-independent model can equal or even outperform models that explicitly incorporate probe properties, such as the model incorporating probe percent bound developed in Chapter Three. This suggests that with a “spiked-in” concentration series targeting as few as 5-10 genes, reliable estimation of target concentration can be achieved for the entire microarray.