Linguistically annotated corpora have nowadays a crucial theoretical as well as applicative role in Natural Language Processing. Italian still lacks such a resource. The paper describes a large scale effort to fill this gap by developing a multi-level annotated corpus, the Italian Syntactic-Semantic Treebank, which represents one of the main actions of an ongoing Italian national project (SI-TAL). The paper provides an overview the annotated corpus, its architecture and composition. For each annotation level, annotation guidelines are briefly discussed, also in the light of underlying motivations and prospective uses of the resource.