Electroencephalogram)EEG( microstate analysis provides a powerful framework for investigating the rapid spatiotemporal dynamics of the brain that underlie cognitive processes. These quasi-steady-state scalp potential configurations reflect transient activations of large-scale neural networks and provide valuable neurophysiological biomarkers for cognitive and clinical research. In this study, we propose a unified microstate-based framework for characterizing task-dependent brain dynamics using high-density EEG recordings. EEG data were collected from 60 healthy young adults during three experimental sessions, including resting states with eyes open and closed, as well as three self-generated cognitive tasks: episodic memory retrieval, mental arithmetic, and musical imagery. The EEG signals were preprocessed using standard filtering, artifact rejection, and independent component analysis. Microstates were extracted from global field power (GFP) peaks using k-means clustering, and task-specific microstate patterns were constructed. Temporal microstate parameters, including mean duration, occurrence rate, temporal coverage, and global explained variance (GEV), were calculated for each microstate class. The effects of cognitive state and recording session on microstate dynamics were assessed using repeated-measures analysis of variance. SVM classification was evaluated using nested leave-one-subject-out cross validation (nested LOOCV).The results revealed significant task-related modulations in multiple microstate parameters (p < 0.001), indicating systematic reorganization of large-scale brain networks under cognitive conditions. Clear differences were observed between memory and mental computation tasks, indicating differential interaction of attentional and default mode networks. In contrast, session-related effects were relatively weak, supporting the temporal stability of microstate patterns. The highest binary accuracy was 85.8% (RBF SVM; eyes-open vs memory).