3D style transfer aims to render a reconstructed 3D scene with the visual appearance of a given artistic reference, while preserving the original content and maintaining consistency across multiple views. Existing approaches often inherit traditional 2D style transfer paradigms, relying on feature statistic matching with limited semantic awareness, which makes it difficult to maintain semantic and stylistic consistency across views. In this work, we propose a novel framework Multi-view Style Distillation for 3D style transfer that performs multi-view stylization via diffusion models and reconstructs stylized 3D scenes using Gaussian Splatting. The first module, Multi-view Semantically Consistent Style Transfer (MSCST), extracts intermediate attention features from diffusion models to guide the generation of stylized multi-view images that are semantically and stylistically consistent across views. To further enhance style strength and stability, we introduce an adaptive iterative refinement strategy that improves robustness without sacrificing geometric coherence. These images are then used to supervise the construction of a stylized 3D scene via Gaussian Splatting, enabling stable and efficient reconstruction. Extensive qualitative and quantitative experiments demonstrate that our approach surpasses state-of-the-art methods in style fidelity, content preservation, and multi-view consistency.