Iterative-based stereo matching methods have achieved remarkable performance in recent years. However, existing methods neglect the critical issue of feature misalignment. This misalignment is induced by inconsistent receptive fields across different scales, degrading both cost volume construction and initial disparity regression. To address these limitations, we propose Iterative Attention-Guided Feature Reconstruction for Stereo Matching (IAFR-Stereo). We design an Attention-Guided Feature Pyramid Reconstruction (AFPR) module that resolves feature misalignment while leveraging attention mechanisms to guide the reconstruction process. Additionally, we design an Attentional Hybrid Feature Fusion (AHFF) module to adaptively integrate discriminative features from multi-frequency domains, enabling enhanced preservation of structural edges and texture details. Extensive experiments show that IAFR-Stereo effectively alleviates feature misalignment, achieving state-of-the-art accuracy and computational efficiency. Notably, our method ranks among the top performers on the widely used KITTI 2012 and KITTI 2015 benchmarks. Moreover, IAFR-Stereo demonstrates strong zero-shot generalization, achieving competitive results on the Middlebury and ETH3D datasets without any fine-tuning.