Second-generation early intervention research typically involves the specification of multivariate relations between interventions, outcomes, and other variables. Moderation and mediation involve variables or sets of variables that influence relations between interventions and outcomes. Following the framework of Baron and Kenny's (1986) seminal paper, this paper differentiates moderation and mediation conceptually and methodologically. Four cases of moderation defined by the scale of predictor and moderator variable scales are described, and several design and statistical issues associated with testing mediation are discussed. Use of structural equation modeling is proposed to address some of the difficulties in testing moderation and mediation effects. A hypothetical early intervention data set is used to discuss and demonstrate the use of structural equation modeling for examining moderation and mediation.