In order to address the "fragmentation" issue in emergency collaboration networks and enhance the effectiveness of emergency management in response to accident disasters，three typical accident disasters were selected. Based on theory of collaborative governance，a "network density-average path distance" analytical framework was constructed using SNA to map the collaborative relationships among various emergency response entities. At the macro level，indicators such as network size，density，shortest path，cohesion，and centralization were employed to portray the overall characteristics of the emergency collaboration networks. At the meso level，cohesive subgroups were analyzed to explore the clustering patterns of emergency collaboration，with the criterion for subgroup division being the professional nature of emergency matters in all three cases. At the micro level，the characteristics of nodes within the networks were evaluated through the triple indices of degree centrality，closeness centrality，and betweenness centrality. The results indicate that emergency collaboration networks of Changsha self-built building collapse，Xiangshui explosion，and Xinjia hotel collapse accidents exhibited tight-collaboration，tight-centralization，and loose-collaboration structures，respectively. It is recommended that the emergency collaboration networks be optimized from three dimensions: clarifying the rights and responsibilities of emergency response stakeholders，adjusting heterogeneous interest demands，strengthening the leadership of core nodes and leveraging their pivotal roles，and promoting the deep embedding of emergency culture to empower endogenous collaborative governance.