A new general methodology for Musical Pattern Discovery is proposed, which tries to mimic the flow of cognitive and sub-cognitive inferences that are processed when hearing a piece of music. A brief survey shows the necessity to handle such perceptual heuristics and to specify perceptual constraints on discoverable structures. For instance, successive notes between patterns should verify a specific property of closeness. A musical pattern class is defined as a set of characteristics that are shared by different pattern occurrences within the score. Moreover, pattern occurrence not only relies on internal sequence properties, but also on external context. Onto the score is build pattern occurrence chains which themselves interface with pattern class chains. Pattern classes may be inter-associated, in order to formalize relations of inclusion or repetition. The implemented algorithm is able to discover pertinent patterns, even when occurrences are, as in everyday music, translated, slightly distorted, slowed or fastened.