A failure to generalize a finding when theoretically-irrelevant procedural changes are made is discussed from the viewpoint of experimental logic ( two experiments showed that the finding that stimulus imagery is more facilitative than response imagery does not generalize to a paired-associate task of considerable economic importance- children learning the sounds of the letters). It is pointed out that the typical experiment is a weak building block for constructing general theory since it only tests a limited null hypothesis- the specific null whose sampling variance is represented in the error term. It is suggested that " generalization test " be substituted for " significance test " to focus E’s attention on his obligation to remove the more serious restrictions on generality in subsequent experiments.