Form-class was considered as a basis for the organization of words in free recall. The experiment employed a definition of clustering which included organizations composed of words of different from-classes as well as those involving words of the same form-class. A high semantic constraint (HSC) group and a low semantic constraint (LSC) group were differentiated on the basis of the proximity of semantically related words to one another in the presentation materials. Recall protocols were scored for all possible 1-, 2-, and 3-word form-class combinations. A comparison of the relative frequencies of the various form-class combinations which occurred in recall with those expected by chance revealed many statistically significant deviations in both HSC and LSC groups. The specific form-class combinations showing deviations above and below chance levels suggested the operation of direct associations among the words and also the operation of more complex grammatical habits.