RandAugment is a practical automated data augmentation method that eliminates the need for a separate search phase, significantly reducing the search space. Unlike previous methods that required optimization for validation accuracy, RandAugment adjusts regularization strength based on model and dataset size. It works uniformly across different tasks and datasets, achieving state-of-the-art results on CIFAR-10/100, SVHN, and ImageNet. On ImageNet, it achieves 85.0% accuracy, a 0.6% improvement over previous methods and 1.0% over baseline augmentation. On object detection, it improves baseline augmentation by 1.0-1.3% and is within 0.3% mAP of AutoAugment on COCO. RandAugment's interpretable hyperparameters allow investigation of data augmentation's role with varying model and dataset sizes. It is trained directly on the target task without a proxy task, making it efficient and adaptable. The method uses a simple grid search to find optimal augmentation policies, reducing computational cost and complexity. RandAugment outperforms previous automated augmentation approaches on multiple benchmarks, demonstrating its effectiveness and efficiency in improving model performance and robustness.RandAugment is a practical automated data augmentation method that eliminates the need for a separate search phase, significantly reducing the search space. Unlike previous methods that required optimization for validation accuracy, RandAugment adjusts regularization strength based on model and dataset size. It works uniformly across different tasks and datasets, achieving state-of-the-art results on CIFAR-10/100, SVHN, and ImageNet. On ImageNet, it achieves 85.0% accuracy, a 0.6% improvement over previous methods and 1.0% over baseline augmentation. On object detection, it improves baseline augmentation by 1.0-1.3% and is within 0.3% mAP of AutoAugment on COCO. RandAugment's interpretable hyperparameters allow investigation of data augmentation's role with varying model and dataset sizes. It is trained directly on the target task without a proxy task, making it efficient and adaptable. The method uses a simple grid search to find optimal augmentation policies, reducing computational cost and complexity. RandAugment outperforms previous automated augmentation approaches on multiple benchmarks, demonstrating its effectiveness and efficiency in improving model performance and robustness.