A method is proposed for selecting the most informative diagnostic symptoms in lifetime consumption monitoring of complex objects. Typically in such cases many symptoms are available and their suitability cannot be evaluated even with a detailed knowledge of object layout and operation. The proposed procedure involves two stages. Preliminary symptom selection is based on the Singular Value Decomposition (SVD) method. Second stage is based on the information content assessment and employs the continuous analogue of Shannon entropy. An example is presented for a steam turbine fluid-flow system.