The paper describes the methods which should be followed when analysing a rotation experiment containing rotations of different length in which all phases of each rotation are not represented, and in which consequently the mean yields for the different rotations of a particular crop are liable to be affected by annual variations in the yield of that crop. Constants are fitted by the method of least squares, the arithmetical method of successive approximation first described by Stevens being used. The estimation of the errors to which the estimates are subject is considered in detail.