Objective: This study aimed to detect the quality of wheat flour quickly and non-destructively. Methods: The online detection system for wheat flour quality was built in the experiment, which mainly includes the flowing four parts: micro near-infrared spectrometer integrated device, production line adjustable speed simulation device, spectrum online acquisition and control software and online modeling and analysis software. Different modeling algorithms were used to establish quantitative analysis models for moisture, ash and gluten, and the robustness of the online quantitative analysis model was optimized. The influence of different experimental conditions on the modeling results was analyzed. Results: Partial least squares regression (PLSR) algorithm is superior to multiple linear regression (MLR) and principal component regression (PCR) algorithm for quantitative analysis of moisture, ash and gluten. The best modeling effect was obtained by using quartz glass to collect wheat flour samples online, and the best ratio of 15 speed/integration time was 6 000 ms. Conclusion: The wheat flour quality online detection system based on near-infrared spectrum analysis technology realize the nondestructive and rapid online detection of wheat flour quality.