This paper proposes a bidirectional multi-scale implicit neural representation (INR) method for image deraining, named NeRD-Rain. The method leverages multi-scale representations to better explore rain streaks and improve image reconstruction quality. Unlike existing Transformer-based methods that rely on single-scale rain appearance, NeRD-Rain uses an end-to-end multi-scale Transformer that incorporates intra-scale implicit neural representations to facilitate rain removal and enhance model robustness in complex scenarios. The method also introduces an inter-scale bidirectional feedback operation to enable richer collaborative representation across different scales. The proposed approach is evaluated on both synthetic and real-world benchmark datasets, demonstrating superior performance compared to state-of-the-art methods. The source code and trained models are available at https://github.com/cschenxiang/NeRD-Rain. The main contributions include the design of an effective multi-scale Transformer to generate high-quality deraining results, the introduction of implicit neural representations to better learn common rain degradation features, and the integration of a simple yet effective bidirectional feedback propagation operation for better feature interaction across scales. Experimental results on both synthetic and real-world benchmarks show that the proposed approach performs favorably against state-of-the-art methods.This paper proposes a bidirectional multi-scale implicit neural representation (INR) method for image deraining, named NeRD-Rain. The method leverages multi-scale representations to better explore rain streaks and improve image reconstruction quality. Unlike existing Transformer-based methods that rely on single-scale rain appearance, NeRD-Rain uses an end-to-end multi-scale Transformer that incorporates intra-scale implicit neural representations to facilitate rain removal and enhance model robustness in complex scenarios. The method also introduces an inter-scale bidirectional feedback operation to enable richer collaborative representation across different scales. The proposed approach is evaluated on both synthetic and real-world benchmark datasets, demonstrating superior performance compared to state-of-the-art methods. The source code and trained models are available at https://github.com/cschenxiang/NeRD-Rain. The main contributions include the design of an effective multi-scale Transformer to generate high-quality deraining results, the introduction of implicit neural representations to better learn common rain degradation features, and the integration of a simple yet effective bidirectional feedback propagation operation for better feature interaction across scales. Experimental results on both synthetic and real-world benchmarks show that the proposed approach performs favorably against state-of-the-art methods.