Celebrity worship has been conceptualized as having pathological and nonpathological forms. To avoid problems associated with item-level factor analysis, 'top-down purification' was used to test the validity of this conceptualization. The respondents (N = 249) completed items modelled after existing celebrity worship questionnaires. A subset of 17 unidimensional and Rasch scalable items was discovered (the local reliability ranged from.71 to.96), which showed no biases related to age and gender. This subset was dubbed the Celebrity Worship Scale (CWS). The items also showed no celebrity bias, indicating that CWS applies equally to acting, music, sports, and 'other' celebrities. The Rasch nature of the items defines celebrity worship as consisting of three qualitatively different stages. Low worship involves individualistic behaviours such as watching and reading about a celebrity. At slightly higher levels, celebrity worship takes on a social character. Lastly, the highest levels are characterized by a mixture of empathy with the celebrity's successes and failures, over-identification with the celebrity, compulsive behaviours, as well as obsession with details of the celebrity's life. Based on these findings, the authors propose a model of celebrity worship based on psychological absorption (leading to delusions of actual relationships with celebrities) and addiction (fostering the need for progressively stronger involvement to feel connected with the celebrity).