Episodic memories are information-rich, often multisensory events that rely on binding different elements [1]. The elements that will constitute a memory episode are processed in specialized but distinct brain modules. The binding of these elements is most likely mediated by fast-acting long-term potentiation (LTP), which relies on the precise timing of neural activity [2]. Theta oscillations in the hippocampus orchestrate such timing as demonstrated by animal studies in vitro [3, 4] and in vivo [5, 6], suggesting a causal role of theta activity for the formation of complex memory episodes, but direct evidence from humans is missing. Here, we show that human episodic memory formation depends on phase synchrony between different sensory cortices at the theta frequency. By modulating the luminance of visual stimuli and the amplitude of auditory stimuli, we directly manipulated the degree of phase synchrony between visual and auditory cortices. Memory for sound-movie associations was significantly better when the stimuli were presented in phase compared to out of phase. This effect was specific to theta (4 Hz) and did not occur in slower (1.7 Hz) or faster (10.5 Hz) frequencies. These findings provide the first direct evidence that episodic memory formation in humans relies on a theta-specific synchronization mechanism.