Analysis of covariance, with a single covariable in a randomized block design, is reconsidered in terms of structural regression. In practice, the covariable is often times uncontrolled, both in the experiment and in repeated sampling, and may follow a linear model as does the variable of interest. If the covariable is unaffected by treatments, then the covariable model may contain a block effect through no treatment effect. When such a model is adequate, the treatment contrast precision can be increased relative to the precision obtained through the standard method of covariance analysis. If, however, treatments affect the covariable, then a treatment effect is included in the model for the covariable. When this covariable model is adequate and when simplifying assumptions are valid, estimates can be given of direct and indirect treatment effects on the variable of interest.