The paper addresses problems resulting from the application of methods, which emphasize homogeneity, in the construction of measures that are expected to represent general upper-level constructs. It distinguishes between homogeneous and semi-homogeneous measures. Whereas homogenous measures allow for one underlying dimension only, semi-homogeneous measures are characterized by the presence of one general and dominating dimension in combination with restricted subordinate dimensions. It is the congeneric model of measurement that tends to create homogeneous measures whereas the c-bifactor model enables the construction of semi-homogeneous measures where the prefix "c" indicates that it is a confirmatory bifactor model. It is made obvious that the successful construction of measures representing upper-level constructs requires the c-bifactor model. The congeneric and c-bifactor models were applied to the social optimism scale since social optimism was known to show a hierarchical structure. As expected, only the c-bifactor model indicated a good model fit.