In this paper, the multi-scale finite element model (FEM) of a composite cable-stayed bridge, Guanhe Bridge, was established based on the Arlequin method firstly. Then a two-step multi-scale FE model updating method was proposed. Furthermore, based on structural health monitoring (SHM) system of Guanhe Bridge, support vector regression (SVR) method was employed to analyse the uncertainty quantification and transmission. It was shown that the errors between the calculated frequencies from the updated multi-scale FEM and the measured frequencies from SHM were less than 3%. In the procedure of inverse uncertainty propagation, the coincidence indexes of the structural parameters were larger than 65%. The deviations between the optimal values of the updated parameters and the corresponding statistical mean values were very small (<5%). Finally, the analysis results indicate that the distributions of the parameters agree well with the assumed normal distribution.