We introduce a novel technique to determine the expression state of a gene from quantitative information measuring its expression. Adopting a productive abstraction from current thinking in molecular biology, we consider two expression states for a gene--Up or Down. We determine this state by using a statistical model that assumes the data behaves as a combination of two biological distributions. Given a cohort of hybridizations, our algorithm predicts, for the single reading, the probability of each gene's being in an Up or a Down state in each hybridization. Using a series of publicly available gene expression data sets, we demonstrate that our algorithm outperforms the prevalent algorithm. We also show that our algorithm can be used in conjunction with expression adjustment techniques to produce a more biologically sound gene-state call. The technique we present here enables a routine update, where the continuously evolving expression level adjustments feed into gene-state calculations. The technique can be applied in almost any multi-sample gene expression experiment, and holds equal promise for protein abundance experiments.