The paper refers to the problem of diagnostic classification of mechanical objects using vibroacoustic symptoms. A new approach based on the rough sets theory is applied to evaluate the symptoms from the point of view of their diagnostic capacity, i.e., the quality of estimation of a technical state of a mechanical object. The approach enables reduction of the set of symptoms to a minimal subset ensuring a satisfactory estimation. The minimal subset is then used to create a classifier of a technical state. Particular attention is paid to a comparison of different methods of calculation of symptom limit values which divide domains of symptoms into intervals corresponding to classes of technical states. The analysed set of data concerns the technical state of rolling bearings installed in a laboratory stand. They are described by a set of symptoms which result from measurements of noise and vibration of bearing housings. The bearings are in good or bad technical states. The paper presents particular steps of the rough sets methodology and gives, as a final result, a classifier of a technical state of bearings based on a minimal subset of symptoms with the greatest diagnostic capacity.