Jackson and Jacobs (1983) argued for three changes in the conduct of message effects research: inclusion of multiple message replications as instances Ofa treatment, recognition ofmessage replications as a source of random variation in the estimation of treatment effects, and attention to issues of message sampling. This review updates their argument and examines 24years of research published in Human Communication Researchfor evidenceof attention to these recommendations. The review shows the following: the prevalence of studies failing to replicate has declined, replications are still rarely recognized as random factors, and researchers who use replications appear to do so for purposes of generalizability and control over confounding but without carefuily analyzing the burden of proof associated with those purposes. An explicitfvamework for discussion of treatment effects in communication is proposed as an advance over the original reasoning of Jackson and Jacobs. sking whether Clinton's "apology" to the American people was effective is a very different question from asking which message A strategies best serve the communicator's aims in making an apology. Experimental communication research occasionally focuses on the former type of question, but mostly it is concerned with the latter type; what is wanted is some sort of abstract description of what works and what does not in some class of communication situations. Questions posed at this more abstract level require a conceptualization of message effects that makes a distinction between a strategic choice understood abstractly and a particular concrete embodiment of that strategic choice in Dale Brashers