We examine how oscillations in the intracranial electroencephalogram (iEEG) relate to human maze learning. Theta- band activity (4-12 Hz in rodents; 4-8 Hz in humans) plays a significant role in memory function in rodents and in humans. Recording intracranially in humans, we have reported task-related, theta-band rhythmic activity in the raw trace during virtual maze learning and during a nonspatial working memory task. Here we analyze oscillations during virtual maze learning across a much broader range of frequencies and analyze their relationship to two task variables relevant to learning. We describe a new algorithm for detecting oscillatory episodes that takes advantage of the high signal-to-noise ratio and high temporal resolution of the iEEG. Accounting for the background power spectrum of the iEEG, the algorithm allows us to directly compare levels of oscillatory activity across frequencies within the 2- to 45-Hz band. We report that while episodes of oscillatory activity are found at various frequencies, most of the rhythmic activity during virtual maze learning occurs within the theta band. Theta oscillations are more prevalent when the task is made more difficult (manipulation of maze length). However, these oscillations do not tend to covary significantly with decision time, a good index of encoding and retrieval operations. In contrast, lower- and higher-frequency oscillations do covary with this variable. These results suggest that while human cortically recorded theta might play a role in encoding, the overall levels of theta oscillations tell us little about the immediate demands on encoding or retrieval. Finally, different patterns of oscillations may reflect distinct underlying aspects of memory function.