This paper investigates audiovisual congruence in virtual reality with both horizontal and vertical offsets between audio and visual rendering. Audiovisual congruence and localization errors are assessed using loudspeaker playback and nonindividualized headphone rendering. To account for the influence of different types of visual information on congruence, presentations of a loudspeaker model and 3D human avatar were compared. Therefore, a new dataset of audiovisual speech was recorded. Results show that human avatar rendering increases perceived congruence, and experienced listeners have an increased tendency to respond with “incongruent” when a loudspeaker model is shown but not when the human avatar is presented. Moreover, a correlation is found between localization precision and audiovisual congruence for horizontally offset stimuli and avatar presentation. For vertical offsets, the angular range of congruence is generally large, and localization errors are high, so no correlation can be observed between the two. The paper contributes congruence ranges for audiovisual speech in virtual reality, which also has implications for augmented reality telepresence use.