To solve the last-mile problem and promote low carbon transportation, hundreds of cities around the world have launched bike-sharing services. A growing number of studies have been mining bike-sharing data from different cities to understand bike-sharing systems. However, there is still a lack of comprehensive understanding of a bike-sharing network structure. Thus, this paper attempts to use complex network theory and spatial autocorrelation analysis approaches to examine structural properties of bike-sharing systems, quantify the importance of bike stations in the network, and evaluate their spatial clustering patterns. This research analyzes bike-sharing trip data from five different sized bike-sharing systems in the United States, Canada, and China. The results of network analysis reveal that the bike-sharing networks have a small-world property with a small average path length and a high clustering coefficient. Compared with medium-scale networks, the connectivity and accessibility of bike stations are relatively low in the large-scale bike-sharing network. Also, the connectivity of bike stations is positively related to their accessibility while they are not highly correlated with their intermediateness. The spatial clustering analysis results indicate that the spatial distributions of connectivity and accessibility of bike stations have a strong global spatial autocorrelation. Also, the stations with high connectivity and accessibility are concentrated in the urban centers, and the stations with low connectivity and accessibility are clustered in the periphery of the urban areas. However, their intermediateness does not show a strong global spatial clustering pattern, which implies that bike-sharing networks consist of multiple sub-groups. The findings provide new insights for transport planners and managers to understand bike-sharing systems and to improve their service quality.