100 patients fulfilling the ICD-10 criteria for depression were rated on the Hamilton Depression Rating Scale. The 1st 17 items of the scale were subjected to principal component analysis with varimax rotation. Five factor solutions were sought corresponding to 6-factor model, 5-factor modeL 4-factor model, 3-factor model and 2-factor model. These factor models were then subjected to confirmatory factor analysis utilizing the structural equation models (path analysis) in order to evaluate the goodness of fit of the yielded models. The 4-factor, 3-factor and 2-factor models appear to be the most appropriate models of the latent structure of the HDRS- 17. The findings suggest that the scale is multidimensional with at least two subscales measuring the severity of the melancholic and nonmelancholic depressions respectively.