The authors explore a testing approach where the concern for selecting the appropriate test input provided to the implementation under test (IUT) is separated as much as possible from the analysis of the observed output. Particular emphasis is placed on the analysis of the observed interactions of the IUT in order to determine whether the observed input/output trace conforms to the IUT's specification. The authors consider this aspect of testing with particular attention to testing of communication protocol implementations. Various distributed test architectures are used for this purpose, where partial input/output traces are observable by local observers at different interfaces. The error-detection power of different test configurations is determined on the basis of the partial trace visible to each local observer and their global knowledge about the applied test case. The automated construction of trace analysis modules from the formal specification of the protocol is also discussed. Different transformations of the protocol specification may be necessary to obtain the reference specification, which can be used by a local or global observer for checking the observed trace. Experience with the construction of an arbiter for the OSI (open systems interconnection) transport protocol is described.>