The optimization of both probe design and analysis algorithms for microarray experiments requires improved understanding and predictability of oligonucleotide hybridization behavior. Our physicochemical theory of GeneChip probe sensitivities divides the probe intensity into an averaged intensity value which serves as a relative measure of the RNA target concentration and the sensitivity of each probe. The sensitivity decomposes into additive terms because of specific and nonspecific hybridization, saturation, the heterogeneous distribution of labels, and intramolecular folding of target and probe. The observed heterogeneity of probe sensitivities is mainly caused by variations of the probe affinity for target binding owing to sequence differences between the probes. The sensitivity values are therefore analyzed in terms of simple molecular characteristics, which consider the base composition and sequence of the probes. We found that the mean sensitivity, averaged over all probes of a chip containing a certain number of bases of one type, strongly increases with an increasing number of C nucleotides per oligomer, whereas A nucleotides show the opposite tendency. These trends are asymmetrical with respect to the number of G and T nucleotides, which have a much weaker, and perhaps a somewhat opposite, effect in probes of intermediate and high sensitivity. The middle base systematically affects the relationship between the sensitivities of perfect match (PM) and mismatch (MM) probes. MM probes are, on the average, more sensitive than the respective PM probes if the middle base is a purine in the PM probe of the respective probe pair. For pyrimidines, this relationship reverses. This purine-pyrimidine asymmetry is partly related to the effect of labeling.