Abstract: This paper exploits recent extentions of H. Simon's notion of modularity, to identify the functional and structural units of an empirical knowledge pattern that define the strongest systematic and self sustaining mechanisms of knowledge transfer and accumulation within the network. These 'core' structures are defined by the connectivity property that every node (technology field) in the core is connected to every other node in the same core by a circular self-sustaining information flow. The core structure achieving the highest rate of knowledge transfer is dominant. Our analysis reconstructs the architecture of the empirical knowledge pattern based on USPTO patent citation data at the level of resolution of 3-digits technology classes, for the period 1975-1999. Based on this fine grained analysis, the issues concerning the relevant notion of modularity, the core structures and their functionality with respect to patent distribution, are addressed. Moreover, the changes through time in the cross-field architecture of knowledge transfer are investigated. The envisaged structural change in the period 1975-1999 offers a few guide lines of interpretation consistent with the idea that the information and communication technologies (ICT), although representing the core of knowledge creation throughout the period, only in the second half became fully integrated with the other sectors.