BACKGROUND: Currently, flour quality evaluation methods are varied, but there are some issues, such as single evaluation indicators and insufficient comprehensiveness. The present study aimed to develop a more comprehensive and rapid evaluation method for flour quality. RESULTS: We first measured nine key quality indicators of dough samples, raw noodle products and cooked noodle products made from wheat flour. Using principal component analysis, we extracted the common information from each indicator and performed dimensionality reduction to establish a comprehensive evaluation index that more comprehensively reflects flour quality. Subsequently, we combined near-infrared spectral technology with chemometric methods to explore the relationship between spectral data and comprehensive evaluation indicators and built a quality prediction model. The accuracy of predictions is crucial for the flour quality prediction model. To improve the model's accuracy, we compared different spectral preprocessing methods and modeling approaches. Finally, we found that the combination of standard normal transformation, full-range wavelength and support vector regression model effectively enhanced the relationship between spectral data and the comprehensive evaluation index. The coefficient of determination of the quality prediction model increased to 0.807 and the relative percent difference reached 5.814 (a coefficient of determination close to 1 and a relative percent difference greater than 2 indicate good predictive capability). The root mean square error for the calibration and prediction sets was reduced to 0.026 and 0.020, respectively. CONCLUSION: This study provides a new method for evaluating wheat flour quality and offers new ideas for the study of other similar issues. © 2025 Society of Chemical Industry.