This paper presents a statistical method for analyzing cDNA microarray expression data using mixed models to assess gene significance. The method allows direct control over the percentage of false positives in gene lists and improves on existing methods by reducing false negatives. It accommodates various experimental designs and can assess significant differences between multiple biological samples. Two interconnected mixed linear models are used to account for variability across and within genes. The method provides a framework for evaluating statistical power and helps researchers select an appropriate number of replicates. It also includes basic graphics for visualizing significant genes. The method is complementary to clustering methods and can be used as a precursor or follow-up to clustering. Replication of spot measurements is essential for determining significance. The method is related to recent work by Kerr et al. and Dudoit et al., and incorporates and extends their best aspects. The method is applied to yeast and lymphoma data, showing its effectiveness in identifying significant genes. The method also addresses the multiple testing problem and provides a way to control the false positive rate. The method is flexible and practical, and can be implemented using commercially available software. The paper also discusses the importance of replication in microarray experiments and the statistical power of different experimental designs. The method is shown to be effective in identifying significant genes in both yeast and lymphoma data. The paper concludes that the method provides a reliable way to assess gene significance in cDNA microarray experiments.This paper presents a statistical method for analyzing cDNA microarray expression data using mixed models to assess gene significance. The method allows direct control over the percentage of false positives in gene lists and improves on existing methods by reducing false negatives. It accommodates various experimental designs and can assess significant differences between multiple biological samples. Two interconnected mixed linear models are used to account for variability across and within genes. The method provides a framework for evaluating statistical power and helps researchers select an appropriate number of replicates. It also includes basic graphics for visualizing significant genes. The method is complementary to clustering methods and can be used as a precursor or follow-up to clustering. Replication of spot measurements is essential for determining significance. The method is related to recent work by Kerr et al. and Dudoit et al., and incorporates and extends their best aspects. The method is applied to yeast and lymphoma data, showing its effectiveness in identifying significant genes. The method also addresses the multiple testing problem and provides a way to control the false positive rate. The method is flexible and practical, and can be implemented using commercially available software. The paper also discusses the importance of replication in microarray experiments and the statistical power of different experimental designs. The method is shown to be effective in identifying significant genes in both yeast and lymphoma data. The paper concludes that the method provides a reliable way to assess gene significance in cDNA microarray experiments.