As potential biomarkers, gene classifiers are gene expression signatures or patterns capable of distinguishing biological samples belonging to different classes or conditions. This is the second of two papers on profiling gene expression in zebrafish (Danio rerio) treated with endocrine-disrupting chemicals of different modes of action, with a focus on comparative analysis of microarray data for gene classifier discovery. Various combinations of gene feature selection/class prediction algorithms were evaluated, with the use of microarray data organized by a chemical stressor or tissue type, for their accuracy in determining the class memberships of independent test samples. Two-way clustering of gene classifiers and treatment conditions offered another alternative to assess the performance of these potential biomarkers. Both gene feature selection methods and class prediction algorithms were shown to be important in identifying successful gene classifiers. The genetic algorithm and support vector machine yielded classifiers with the best prediction accuracy, regardless of sample size, nature of class prediction, and data complexity. A chemical stressor significantly altering the expression of a greater number of genes tended to generate gene classifiers with better performance. All combinations of gene feature selection/class prediction algorithms performed similarly well with data of high signal to noise ratio. Gene classifier discovery and application on the basis of individual sampling and sample data pooling, respectively, were found to enhance class predictions. Gene expression profiles of the top gene classifiers, identified from both microarray and quantitative polymerase chain reaction assays, displayed greater similarity between fadrozole and 17beta-trenbolone than either one to 17alpha-ethinylestradiol. These gene classifiers could serve as potential biomarkers of exposure to specific classes of endocrine disruptors.