Contrastive learning can assist sequential recommendation in improving the performance of predicting the next-interacted item, often by mining self-supervised signals in augmented sequences. However, the noise in the original interaction sequences and the augmented sequences can cause interference in the accurate recommendation. Therefore, in this paper, we propose a robust interest-guided contrastive sequential recommendation model, named RINo4CSR, which improves the performance by identifying and eliminating the noise in user interaction sequences. Specifically, RINo4CSR learns interest prototypes with respect to overall users by clustering items. Then, RINo4CSR utilizes augmented sequences to construct a support set for each user and extracts the user’s robust interests reflected by his/her support set, thus bypassing the interference of the interaction noise with the user interest modeling. Further, guided by the user’s interests, RINo4CSR not only detects and removes noise of the support set; but also captures two types of self-supervised signals. The experimental results on three public datasets show that our model outperforms existing sequential recommendation models in terms of hit ratio and NDCG and can provide improved robustness of performance for users with noisy interactions.