In several paired-associate (PA) tasks, triads of words served as stimuli. The associative relationship between the words of a triad was systematically varied. Words of each H (horizontal) triad were interassociated, while those of the I (independent) triads were not. Words of the V (vertical) triads were associated with words of other V triads but not with each other. In Experiment I, 30 Ss learned each type of item. Performance on the H items was facilitated, and that on the V items was hindered (p < .001). Experiment II (N = 30) replicated this result and provided control data with single-word stimuli. PA performance with stimulus triads was always poorer than that with single-word stimuli (p < .001). Experiment III (N = 60) examined the generality of these results. A single view of PA learning is proposed to integrate several interpretations of the results, and implications for serial learning are discussed.This article reports three experiments that examine the effect of compound stimuli on paired-associate (PA) learning. Two of these experiments show that PA learning is facilitated when the ingredients of the compound stimulus are interassociated. The third experiment, however, limits the generality of this conclusion. These experiments provide empirical data that may help clarify the nature of the stimulus in serial learning.The "chaining hypothesis" of serial learning asserts that each item of a serial list comes to elicit the following item, but a number of direct experimental tests (e.g., Primoff, 1938;Young, 1959) have disconfirmed this hypothesis. In these experiments a serial list and a PA list were prepared such that each pair of adjacent items of the serial list became a pair of associates in the PA list; S then learned both lists in either order. Hereafter, this PA task is called a "chained PA task."According to the chaining hy-