Marascuilo and Levin's (1970) notion of Type IV errors is extended, with respect to the interpretation of interactions in analysis of variance (ANOVA) designs. To help clarity what an interaction is and what it is not, in terms of the ANOVA model, the following points are made: (i) interactions should be thought of as linear contrasts involving particular cell means; (ii) such contrasts may be both specified and directional, in that they may be defined to test an investigator's a priori hypotheses; and (iii) even the layman's conceptualization of interactions fits nicely into the ANOVA model. (Author)