Generalized recognition of the kernel, passive, negative, and passive-negative transformations of a number of sentences was investigated. A significant amount of generalization was obtained within a transformationally-defined sentence family relative to the generalization obtained between sentence families when subjects were requested to register recognition of sentences previously read. A distance measure was used to compare generalization decrements between various constructions. The distance between sentences related by the passive transformation was less than between those related by the negative or a combination of negative and passive transformations. The distance between sentences related by negative transformations was not consistently less than that related by the combination of passive and negative. When the order of transformation was considered, the distance between sentences related by both passive and negative transformations equaled the sum of the distance between sentences related by the passive and negative. A satisfactory spatial representation of this data was Attneave's "city block" model. (KB)