Synchronization of dispersed brain areas comprising the Default Mode Network (DMN) has been implicated in a range of cognitive processes (e.g., episodic memory retrieval, dream recall, creativity) [1,2]. Here, we probe the role of the DMN in divergent thinking through intracranial recordings in 8 patients with electrodes implanted for epilepsy monitoring. Subjects performed a creativity-requiring alternate uses task (AUT) and two control tasks (Fig. 1a)—a basic arithmetic task (BAT) and an object characterization task (OCT). The originality of subjects’ AUT responses was automatically scored using natural language processing techniques. First, we uncovered a relationship between neural activity patterns and creativity levels, characterized by increased high gamma (Fig. 1b, left) and decreased theta power (right) across DMN nodes in high-creative compared to low-creative states. High gamma activity is thought to reflect local neuronal population firing [3-7]—hence, our result reveals state-specific DMN recruitment during highly creative states. We next examined the temporal dynamics of DMN engagement in the creativity task as well as the control tasks (Fig. 1c). During the stimulus presentation window of AUT, high gamma power initially peaked before returning to baseline; during the response window, high gamma activity exhibited an even greater increase. A similar pattern was observed for OCT, but with lower response-period high gamma modulation. Meanwhile, no high gamma modulation occurred in BAT (or, if anything, it slightly decreased). Theta power, by contrast, uniquely decreased during the stimulus period of the creativity task (whereas both control tasks were characterized by increased theta power throughout the stimulus presentation). At the end of the stimulus, AUT theta returned to baseline; OCT and BAT theta remained elevated. Upon entering the response window, theta power declined in all trial types, nearing baseline levels for BAT and falling below baseline for AUT and OCT (with AUT still displaying the lowest theta power). Finally, using a linear support vector machine classifier to decode trial type from the electrophysiology data, we were able to successfully distinguish trials involving divergent thinking vs. mathematical reasoning (Fig. 1d). This novel finding suggests that divergent thinking depends on a unique network-wide pattern of neural activity and that distinct “brain states” underlie various modes of thought.