A complex diagnostic object typically generates a large number of diagnostic symptoms. For proper lifetime consumption monitoring it is necessary to select those which best represent technical condition deterioration. Such selection may be based on the Singular Value Decomposition method. The paper presents an alternative approach, which employs an information content measure. As the end of object life is approached, symptoms are to an increasing extent dominated by deterministic lifetime consumption processes and therefore become more predictable. Thus symptoms with the highest information content decrease rate should be considered most useful. In proper assessment, however, symptom sensitivity to condition parameters should also be addressed. A measure referred to as representativeness factor is proposed. Suitability of such approach is demonstrated for the fluid-flow system of a large steam turbine.