The general linear model (GLM) captures the relationship between one or more response variables , and several design variables . One postulates a relationship between and which involves parameters , :

where represents the error. Typically, we assume that has mean , i.e. that the error is unbiased.
Variables could be interval or ratio measurements (such as mass, time, temperature etc.), or can be indicator variables. An indicator variable has value of or . A set of indicator variables may be assigned to indicate the membership of a unit to a particular treatment group. Thus, if there are treatment groups, we would have indicator variables. The -th indicator variable of the set would have value of for all units in group and otherwise.
In the meat packaging example there are 4 treatment labels, and thus there are 4 indicator variables -. The GLM for this example would be (assuming there is no ):

If we restrict the model to the -th treatment group, we can write:

which is equivalent to the full model introduced before (formally setting ). Similarly, we can write the reduced model as a GLM:

The parameters of the GLM are estimated using the least squares method.