Two experiments examined the relevance of frequency theory and feature analysis models of verbal discrimination learning (VDL) for differential eyelid conditioning to related words. The single reinforced and nonreinforced stimuli were: (a) high associates, (b) homonyms, (c) members of the same taxonomic class, (d) unrelated nouns, or (e) unrelated noun and adjective or verb items. Experiment I examined these five types of relatedness at 1000-msec and 14oo-msec interstimulus intervals (lSI), while Experiment II tested ISIs of 1000 msec and 600 msec. Conditioned discrimination was generally poorer at the 6oo-msec lSI, but neither VDL model's relatedness interference predictions were fully confirmed. Contrary to frequency theory, associates were no more difficult to discriminate than unrelated words. The feature analysis model was partially supported, in that taxonomic but not homonymic relatedness produced some interference at the 606-msec lSI. Possible process differences between verbal recognition learning and differential conditioning which may account for these results were discussed.The present research examined the relevance of recognition learning theory for human classical conditioning. Specifically, differential conditioning to verbal conditioned stimuli (CSs) was viewed as an analog to verbal discrimination learning (VDL), and the predictions of two alternative VDL theories were contrasted within the context of differential eyelid conditioning to related verbal discriminanda. The positive stimulus word (CS+), reinforced by occurrence of the unconditioned stimulus (US), and the negative or nonreinforced word (CS-) were either high normative associates, members of the same taxonomic category, homonyms, unrelated nouns, or unrelated noun-non noun (adjective or verb) combinations.According to frequency theory (Ekstrand, Wallace, & Underwood, 1966), associative relatedness between correct (R) and incorrect (W) items interferes with VDL because the occurrence of the R and W items as implicit associative responses (lARs) retards the development of discriminable subjective R-W frequency differences. The analogous prediction in the present situation is that differential conditioning should be detrimentally affected by associative relatedness, but not by homonymic or taxonomic relatedness, assuming