To fully explore the correlation and temporal characteristics between online monitoring data of dissolved gases in oil-immersed transformers, and to improve the accuracy of early defect warnings and defect type identification, a method based on Jensen-Shannon divergence and dynamical network marker is proposed for early transformer defect warning and identification. First, the monitored quantities of dissolved gases in transformer oil are mapped as nodes of a complex network reflecting transformer state evolution, allowing the analysis of transformer deterioration based on the time-series monitoring data. Then, prediction models for the components of dissolved gases are established using historical health condition data from the transformer’s oil chromatography online monitoring system. Jensen-Shannon divergence is introduced to construct an inconsistency indicator, quantifying the dynamic difference between actual monitored values and predicted values under healthy conditions. This helps identify critical dynamical network marker during defect-prone states. Case studies show that this method can not only provide accurate early warnings for transformer defects but also effectively identify defect types.