The paper presents an attempt to match signal processing derived features to the higher-level (linguistic) description of timbre in order to support the timbre driven retrieval of violins. The experiments are performed on a set of 53 instruments recorded during the 10th Henryk Wieniawski International Violinmakers Competition held in Pozna.., Poland, in 2001. The results add to the recent research approach to apply computational intelligence to bridge the semantic and conceptual gap, which exists between the low level description and higher-level interpretation of multimedia. The taxonomy used for violin voices is correlated with objective attributes of entirely distinct character so that semantically disjoint categories are created, providing supportive cues for further analysis. The results should deliver new, enriching aspect to the design of methodology of inferring preference models from the objective characteristics of digital media.