We examined the visual perception of joint actions, in which two individuals coordinate their body movements in space and time to achieve a joint goal. Animations of interacting action pairs (partners in human interactions) and non-interacting action pairs (individual actors sampled from different interaction sequences) were shown in the experiment. Participants were asked to rate how likely the two actors were interacting. The rating data were then analyzed using multidimensional scaling to recover a two-dimensional psychological space for representing joint actions. A descriptive model based on ordinal logit regression with a sparseness constraint was developed to account for human judgments by identifying critical features that signal joint actions. We found that identification of joint actions could be accomplished by assessing inter-actor correlations between motion features derived from body movements of individual actions. These critical features may enable rapid detection of meaningful inter-personal interactions in complex scenes.