The ability to detect outliers in a scatterplot is elemental for understanding and interpreting the underlying data, especially given the high prevalence of scatterplots in analytical visualizations of all kinds. The aim of this thesis is to find out how participants perceive and detect outliers in scatterplots. This was done by conducting a user study and collecting both quantitative and qualitative data, with the goal of finding patterns in outlier definitions. For this, participants viewed synthetic scatterplots of varying point count and correlation coefficient, and were asked to mark all points they deemed to be outliers. The results were analyzed by comparing the marked outliers to commonly accepted statistical outlier definitions, and categorizing the subjective outlier definitions offered by the participants. I have found that point count and correlation coefficient have an effect on the selected outlier count and necessary time, although there is a shared preferred range for the outlier count that is consistent across many participants. Additionally, I discovered that participants have preferences regarding outlier metrics, with a geometric shape of a point distribution being valued the highest by almost all participants, while other preferences seemed more personal and varied between participants. Overall similarity between certain statistical outlier metrics and the subjective outlier selection could also be noticed in a majority of cases, however there are some exceptions that need to be studied further.