Observation of other people's choices can provide useful information in many circumstances. However, individuals may not utilize this information efficiently, i.e., they may make decision-making errors in social interactions. In this paper, I use a simple and transparent experimental setting to identify these errors. In a within-subject design, I first show that subjects exhibit a higher level of irrationality in the presence than in the absence of social interaction, even when they receive informationally equivalent signals across the two conditions. A series of treatments aimed at identifying mechanisms suggests that a decision maker is often uncertain about the behavior of other people so that she has difficulty in inferring the information contained in others' choices. Building upon these reduced-from results, I then introduce a general decision-making process to highlight three sources of error in decision-making under social interactions. This model is non-parametrically estimated and sheds light on what variation in the data identifies which error.