Reverberation plays a fundamental role in auralization of enclosed spaces as it helps to produce more realistic and immersive virtual 3D sound scenes. However, rigorous simulation of room acoustics can be rather computationally expensive, and in practice it is common to use simplified models, sometimes at the cost of spatial accuracy. In the present study, we carry out a comparison between different techniques, and explore the trade-off between computational cost and perceived quality when rendering reverberation in the binaural domain.In a previous study, several binaural reverberation techniques of varying complexity were compared in a subjective test. An approach based on Ambisonics and virtual loudspeakers was used, where reverberation was simulated by means of acoustic modelling software (ray-tracing) and then rendered through different Ambisonic orders (first and third) and loudspeakers configurations (stereo, tetrahedron, dodecahedron). Additionally, two simpler methods were considered: “mono diotic”, where reverberation was presented diotically, and “mono panned”, where reverberation for each sound source was spatialised at the same position as the direct sound. For all test conditions, the direct sound path was rendered through convolution with the Head Related Impulse Response (HRIR), while the reverberation was generated from simulations where the direct path was removed. Therefore, differences between test conditions referred exclusively to reverberation. Convolutions were performed offline and assuming a static listener. A listening test (N = 75) consisting on blind A/B comparisons was conducted remotely using a web-based application.Two versions of the study were designed in order to test different types of scenarios - single/multiple sources, moving/static sources - and audio content - speech/music. In the first version, participants listened to a simulation of a person walking in circles around them, which was rendered as two moving sources (speech and footsteps) spatialised at different elevations. The second version simulated a concert, where the listener faced three instruments distributed along the frontal hemifield: a drum kit, which was rendered as three independent static sources, a saxophone and a piano, both of which were rendered as individual static sources.Initial results show that for the “speech” content, all the proposed reverberation methods obtained similar results in terms of realism and spatialisation accuracy, suggesting that complex renderings, such as the 20-loudspeaker 3rd order Ambisonic are perceived as similar to simpler renderings, such as stereo or “mono diotic”. The only exception was the “mono panned” condition, which obtained statistically significant lower scores overall. Interestingly, recently collected data for the “music” condition showed very similar results, except for the “mono panned” condition, which obtained statistically significant higher scores in terms of spatial impression. The implications of these seemingly contradictory outcomes are currently being investigated. It is worth noting that the online nature of the study did not allow for the control of environmental conditions e.g. noise level, quality of audio reproduction equipment and reproduction level. To further explore the results observed so far, a third study will be carried out in controlled laboratory conditions. Other proposed modifications in the test include the introduction of head tracking and the use of acoustic measurements to generate the reverberation, instead of geometrical acoustic simulations. Additionally, an extra condition based on Spatial Decomposition Method for Room Impulse Responses will be introduced as a reference condition. The details of the study and preliminary results will be presented at the conference.