Abstract: In the process of road construction and maintenance，the efficient and accurate measurement of aggregate gradation in asphalt mixture is an important factor to ensure the stability of mixture skeleton structure and construc-tion quality． Considering the methods based on a single geometric model can not meet the requirements of particle size calculation accuracy in construction practice，a neural network model for road aggregate size calculation based on multiple features was proposed． Firstly，geometric features were extracted from the collected aggregate particle images，and the extracted feature data were cleaned and normalized to establish the sample data set． Secondly，the characteristic factors with strong correlation with aggregate particle size were extracted by correlation analysis． Fi-nally，multi-layer perceptron (MLP) neural network was constructed to train the data set，and the important cha-racteristics weight representing aggregate particle size was obtained by the sensitivity analysis． Thus the particle size of coarse aggregate can be accurately calculated． The results show that the aggregate particle size calculation meth-od proposed in this paper has a higher fitting accuracy (R² =0. 91) than the results measured by the traditional geo-metric models such as secondary moment and equivalent ellipse． It not only improves the accuracy obviously，but also realizes fast virtual screening and significantly improves the subsequent screening efficiency．

Key words: aggregate size, multiple feature factor, geometric feature, correlation analysis, virtual screening, multi-layer perceptron neural network