An important mental health concern arising from the COVID-19 pandemic was the loneliness resulting from the lockdown measures taken by many countries due to the outbreak. Thus, loneliness needs to be studied in detail for intervention purposes. The UCLA Loneliness Scale is the most widely used measure of loneliness. However, the 20-item measure has some drawbacks, including questionnaire fatigue, especially when used with other measures, and a lack of accurate responses to all items. In this study, we evaluated two short forms of the UCLA Loneliness Scale used in other studies (UCLA-10 and UCLA-8), as well as a five-item version developed in this study, through the classical test theory and Mokken analysis with two different sample groups (teachers, N = 337; students, N = 355). The results of the Mokken analysis and the confirmatory factor analysis showed that the UCLA-10 and UCLA-5 were unidimensional and exhibited reliability and convergent validity. In contrast, the UCLA-8 was multidimensional and violated several assumptions of monotonicity and invariant item-ordering, although it showed satisfactory reliability and concurrent validity. All three versions of the scale correlated significantly with other indices of psychological well-being regarding concurrent validity, and the correlation coefficients were comparable to the coefficients of the 20-item version with the same variables. All three versions of the loneliness scale also demonstrated satisfactory convergent and discriminant validity. The results indicated that the UCLA-10 and UCLA-5 can be used as reliable and valid measures when the use of the long version of the UCLA Loneliness Scale is unsuitable.