Factorial experiments are described and the importance of interactions emphasized with suggestions about how they should be interpreted. A distinction is made between factors introduced only to see if they will provoke an interaction and those actually under study. Split-plot designs receive special attention. Factorial experiments often involve a large number of treatments and ordinary block designs may be ineffective in controlling environmental variation. If factors have few levels, as in exploratory experiments, the usual device for reducing block size is confounding, which is explained along with single-replicate experiments, partial replication and hidden replication. Alternatively, non-orthogonal designs and analysis of data by nearest-neighbour and spatial methods might prove useful. The need for randomization and the role of significance are discussed. It is pointed out that interactions can sometimes be avoided by transformation of the variate.