Genes in a cell are not independent. They interact with one another to complete certain biological processes or implement certain molecular functions. Genes' dependency is encoded in biological pathways or gene networks. In this talk, I will present some Markov random field (MRF) models to account for gene regulatory dependency in analysis of Microarray gene expression data. These MRF-based models can efficiently utilize the known gene network structures in identifying genes and sub-networks that are related to diseases. Simulation results and applications to real examples demonstrate that our approach has better performance than the methods not utilizing any biological information.