Starting from the pioneering works on software architecture precious guidelines have emerged to indicate how computer programs should be organized. For example the "separation of concerns" suggests to split a program into modules that overlap in functionality as little as possible. However these recommendations are mainly conceptual and are thus hard to express in a quantitative form. Hence software architecture relies on the individual experience and skill of the designers rather than on quantitative laws. In this article I apply the methods developed for the classification of information on the World-Wide-Web to study the organization of Open Source programs in an attempt to establish the statistical laws governing software architecture.