ROCR is an R package for evaluating and visualizing the performance of scoring classifiers. It provides over 25 performance measures that can be combined to create two-dimensional performance curves. It includes standard methods for investigating trade-offs between performance measures, such as ROC graphs, precision/recall plots, lift charts, and cost curves. ROCR integrates with R's graphics capabilities, allowing for highly adjustable plots. It is easy to use, with only three commands and reasonable default values for optional parameters. ROCR is useful for evaluating scoring classifiers in bioinformatics, where predicting phenotypic properties of HIV-1 from genotypic information is important. It is also used in other biological problems, such as microarray analysis, protein structural and functional characterization, genome annotation, protein-ligand interactions, and structure-activity relationships. However, predictive modeling is complicated by class skew, class-specific misclassification costs, and noise from experimental variability. The real-valued output of scoring classifiers is converted into a binary class decision by choosing a cutoff. Since no cutoff is optimal for all performance criteria, cutoff choice involves a trade-off among different measures. Trade-offs between two criteria are visualized as cutoff-parametrized curves. Examples include ROC graphs, sensitivity/specificity curves, lift charts, and precision/recall plots. ROCR allows for creating cutoff-parametrized performance curves by combining two of more than 25 performance measures. Curves from different validation runs can be averaged, and standard deviations, standard errors, and box plots are available to summarize variability. The parametrization can be visualized by printing cutoff values or coloring the curve according to the cutoff. All components of a performance plot are adjustable. ROCR is a comprehensive tool for evaluating scoring classifiers and producing publication-quality figures. It allows for studying the intricacies of biological datasets and their implications on classifier performance.ROCR is an R package for evaluating and visualizing the performance of scoring classifiers. It provides over 25 performance measures that can be combined to create two-dimensional performance curves. It includes standard methods for investigating trade-offs between performance measures, such as ROC graphs, precision/recall plots, lift charts, and cost curves. ROCR integrates with R's graphics capabilities, allowing for highly adjustable plots. It is easy to use, with only three commands and reasonable default values for optional parameters. ROCR is useful for evaluating scoring classifiers in bioinformatics, where predicting phenotypic properties of HIV-1 from genotypic information is important. It is also used in other biological problems, such as microarray analysis, protein structural and functional characterization, genome annotation, protein-ligand interactions, and structure-activity relationships. However, predictive modeling is complicated by class skew, class-specific misclassification costs, and noise from experimental variability. The real-valued output of scoring classifiers is converted into a binary class decision by choosing a cutoff. Since no cutoff is optimal for all performance criteria, cutoff choice involves a trade-off among different measures. Trade-offs between two criteria are visualized as cutoff-parametrized curves. Examples include ROC graphs, sensitivity/specificity curves, lift charts, and precision/recall plots. ROCR allows for creating cutoff-parametrized performance curves by combining two of more than 25 performance measures. Curves from different validation runs can be averaged, and standard deviations, standard errors, and box plots are available to summarize variability. The parametrization can be visualized by printing cutoff values or coloring the curve according to the cutoff. All components of a performance plot are adjustable. ROCR is a comprehensive tool for evaluating scoring classifiers and producing publication-quality figures. It allows for studying the intricacies of biological datasets and their implications on classifier performance.