This paper reports the results of manipulations of word features for the magnitude of priming effects. In Experiment 1, the printed frequency of the target words and the number of connections among their associates were varied, and during testing participants were given cues and asked to produce the first word to come to mind as rapidly as possible in implicit free association. Priming effects were greater for lowfrequency words and for those with many connections among their associates.In Experiments 2 and 3, target words were presented under incidental or intentional learning conditions during study, and the presence of direct preexisting connections from targetto cue and from cue to targetwas varied. Priming effects were greater when either connection was present, with each connection having additive effects. In Experiments 4 and 5, priming effects for indirect links (shared associatesand mediators) were examined. The results of these experiments indicate that priming in free association depends on both the general accessibility of the target as a response and the strengthening of direct target-to-cue connections. These findings raise problems for theories that attribute priming only to target accessibility or only to target-to-cue association.