Handwritten mathematical expression generation (HMEG) is a challenging task due to the complex structure and significant symbol distortion. In addition, the wide variety of handwriting styles, e.g., variations in symbol size and slant, aggravates the difficulties of the HMEG. In this paper, we propose a novel Visual-aware Multi-Representation Fusion Network (VMF-Net) to tackle the challenging HMEG task. Specifically, we introduce a Multi-Representation Enhanced Generator (MEG) that guides the generation process by various visual attributes, such as high-frequency components and expression prototypes, in conjunction with structural features. Additionally, we design a Progressive Fusion Module (PFM) to capture interactions between visual features, while an Adaptive Filter (AF) suppresses redundant style features. Extensive experiments on widely used benchmark datasets demonstrate that our VMF-Net is an effective HMEG model and outperforms state-of-the-art models remarkably.