Abstract: High-quality public travel demand has become one of the major challenges for Intelligent Transportation Systems. For traffic demand prediction, most of existing models adopted graphs with fixed structure to describe the spatial correlation of traffic demand, ig-noring that traffic demand has different spatial dependence at different scales. Thus, a Multi-scale Spatial-temporal Graph Convolutional Network (MSTGCN) model was proposed for traffic demand prediction. Firstly, global demand similarity graph and local demand simi-larity graph were constructed at global and local scales. Two graphs can capture long-term stable and short-term dynamic features of travel demand. Graph Convolutional Networks were introduced to extract global and local spatial information in two graphs. Besides, MSTGCN adopted attention mechanism to combine two kinds of spatial information adaptively. Moreover, gated recurrent unit was used to capture time-varying features of travel demand. Taking New York Citi (NYC) Bike and NYC Taxi as samples, experimental results show that Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Pearson Correlation Coefficient (PCC) of MSTGCN were 2.7886, 1.7371, and 0.7992 on NYC Bike. And they were 9.5734, 5.8612, and 0.9631 on NYC Taxi. It proves that MSTGCN can effectively mine multi-scale spatial-temporal features of travel demand so as to accurately predict future travel demand.