The aim of this study was to present the dilemmas in the diagnosis of eating disorders on an example of division of a group of girls with diagnosed eating disorders on the basis of the self-assessment questionnaire results which provides an alternative to the clinical criteria. The study covered 116 girls diagnosed with any of the eating disorders according to DSM-IV consulted for the first time between 2002 and 2004 in the outpatient clinic of the Child and Adolescent Psychiatry Unit, University Hospital in Cracow, Poland. Due to statistical limitations of the model only dependent variables such as self-assessment of the presence of depressive symptoms (measured by Beck Depression Questionnaire, BDI), eating attitudes (measured using Eating Attitudes Test, EAT26), self-image (Social Self from Offer’s Self-Image Questionnaire, JaSpoł QSIA) and family relations (Intimacy scale from the Family of Origin Scale, INT FOS) were subjected to k-means cluster analysis. The conducted analyses led to the emergence of five clusters which did not correspond to the division into particular clinical diagnoses. All the clusters turned out to differ from one other in the context of the results of their constituent questionnaire scales. Cluster One proved to characterise individuals with self-assessed low severity of eating problems and depression and a favourable image of their social functioning and family relations. The interaction observed in Cluster Five was the opposite. The majority of the patients in Cluster One were diagnosed with restrictive anorexia, while in Cluster Five with bulimia. Cluster Three turned out to be most different from the others. Positive self-image and family relations were accompanied by highself-assessed severity of eating problems and low depression. The obtained results indicate that the clinical division of eating disorders proposed in ICD and DSM is not the only possible. Other ways of grouping patients may be of interest not only cognitively but may also have some clinical value.