The paper analyses the problem of aggregation of internal quality characteristics of an information system specification language. It surveys main categories of aggregation operators, their mathematical and behavioural properties and proposes how to aggregate quality characteristics taking into account measurements of all lowest-level characteristics in a proper way. The paper shows that four different kinds of dependencies can occur among characteristics of internal quality and discusses in detail what aggregation techniques are reasonable for each case. It also describes a heuristic for minimisation of possible deviations of measurement results obtained after using a number of different metrics or because of the inaccuracy of used measurement techniques.