Time-series microarray data can capture dynamic genomic behavior not available in steady-state expression data, which has made time-series analysis especially useful in the study of dynamic cellular processes such as the circadian rhythm, disease progression, drug response, and the cell cycle. Using the information available in the time-series data, we address three related computational problems: the prediction of gene expression levels from previous time steps, the simulation of an entire time-series microarray dataset, and the reconstruction of gene regulatory networks. We model the gene expression levels using a linear model, due to its simplicity and the ability to interpret the coefficients as interactions in the underlying regulatory network. A stepwise multiple linear regression method is applied to fit the parameters of the linear model to a given training dataset. The learned model is utilized in predicting and replicating the time course of the expression levels and in identifying the regulatory interactions. Each predicted interaction is also associated with a statistical significance to provide a confidence measure that can guide prioritization in further costly manual or experimental verification. We demonstrate the performance of our approach on several yeast cell-cycle datasets and show that it provides comparable or greater accuracy than existing methods and provides additional quantitative information about the interactions not available from the other methods.