Experimental studies are often employed to test the effectiveness of teaching innovations such as new pedagogy, curriculum, or learning resources. This article offers guidance on good practice in developing research designs, and in drawing conclusions from published reports. Random control trials potentially support the use of statistical inference, but face a number of potential threats to validity. Research in educational contexts often employs quasi-experiments or natural experiments rather than true experiments, and these types of designs raise additional questions about the equivalence between experimental and control groups and the potential influence of confounding variables. Where it is impractical for experimental studies to employ samples that fully reflect diverse populations, generalisation is limited. Series of small-scale replication studies may be useful here, especially if these are conceptualised as being akin to multiple case studies, and complemented by qualitative studies. Control conditions for experimental studies need to be carefully selected to provide the most appropriate test for a particular intervention, and considering the interests of all participants. Control groups in studies that replicate innovations that have been widely shown to be effective in other settings should experience teaching conditions that reflect good practice and meet expected teaching standards in the research context.