Musical schemata theory entails the classification of subphrase-length progressions in melodic, harmonic and metric feature-sets as named entities (e.g., `Romanesca', `Meyer', `Cadence', etc.), where a musical schema is characterized by factors such as music content and form, position and tonal function within phrase structure, and interrelation with other schemata. To examine and automate the task of musical schemata classification, we developed a novel musical schemata classifier. First, we tested methods for exact and approximate matching of user-defined schemata prototypes, to establish the notions of identity and similarity between composite music patterns. Next, we examined methods for schemata prototype extraction from collections of same-labelled annotated examples, performing training and testing sessions similar to supervised learning approaches. The performance of the above tasks was verified using the same annotated dataset of 40 keyboard sonata excerpts from pre-Classical and Classical periods. Our evaluation of the classifier sheds light on: (a)~ability to parse and interpret music information, (b)~similarity methods for composite music patterns, (c)~categorization methods for polyphonic music.