Many Internet information-management applications (e.g., information integration systems) require a library of wrappers, specialized information extraction procedures that translate a source's native format into a structured representation suitable for further application-specific processing. Maintaining wrappers is tedious and error-prone, because the formatting regularities on which wrappers rely change frequently on the decentralized and dynamic Internet. The wrapper verification problem is to determine whether a wrapper is operating correctly. Standard regression testing approaches are inappropriate, because both the formatting regularities on which wrappers rely and the source's underlying content may change. We introduce RAPTURE, a fully-implemented, domain-independent wrapper verification algorithm. RAPTURE computes a probabilistic similarity measure between a wrapper's expected and observed output, where similarity is defined in terms of simple numeric features (e.g., the length, or the fraction of punctuation characters) of the extracted strings. Experiments with numerous actual Internet sources demostrate that RAPTURE performs substantially better than standard regression testing.