The rapid increase in semistructured data usage has lead to the development of various database systems for semistructured data. Web services and applications that utilize large amounts of semistructured data require data to remain consistent and be stored efficient. Several normalization algorithms for semistructured database systems have been developed to satisfy these needs. However, these algorithms lack the verification that would ensure that data and constraints among the data are not lost or corrupted during normalization. In this paper, we propose a set of correctness criteria for normalization of semistructured data, which require that functional dependencies are preserved, data is not lost, and spurious data is not created during normalization. We use the Z specification language to provide a precise and declarative definition of our criteria.