We propose a framework using contrastive learning as a pre-training task to\nperform image classification in the presence of noisy labels. Recent strategies\nsuch as pseudo-labeling, sample selection with Gaussian Mixture models,\nweighted supervised contrastive learning have been combined into a fine-tuning\nphase following the pre-training. This paper provides an extensive empirical\nstudy showing that a preliminary contrastive learning step brings a significant\ngain in performance when using different loss functions: non-robust, robust,\nand early-learning regularized. Our experiments performed on standard\nbenchmarks and real-world datasets demonstrate that: i) the contrastive\npre-training increases the robustness of any loss function to noisy labels and\nii) the additional fine-tuning phase can further improve accuracy but at the\ncost of additional complexity.\n