Increased reliance on optimistic data replication has led to burgeoning interest in tools and frameworks for synchronizing disconnected updates to replicated data. To better understand the issues underlying the design of generic and heterogeneous synchronizers, we have implemented an experimental framework, called Harmony, that can be used to build synchronizers for tree-structured data stored in a variety of concrete formats. We present Harmony’s architecture, formalize its key components (a simple core synchronization algorithm together with a set of user-defined mappings between diverse concrete data formats and common abstract schemas suitable for synchronization), and discuss how the framework can be used to synchronize a variety of specific types of application data by suitable encodings into trees—including sets, records, tuples, relations, and, with some limitations, lists and ordered XML data.