The experiment reported on here investigating the syllabic attachment of post-vocalic nasals and l grew out of an earlier experiment which compared the post-vocalic liquids r and l (Laubstein 1989). Using the error elicitation technique described in Motley et al (1983), induced errors involving Vr and Vl sequences were compared. The two sequences behaved quite differently, suggesting different syllabic attachments for the two liquids. Vowel-r sequences consistently interacted with single vowels or vowel glide sequences and such sequences never split up, supporting the hypothesis (based on naturally occurring errors) that vowel-r sequences belong to a uni-positional peak constituent and supporting Kahn’s argument that English r, like w and y, is a glide (Kahn 1976). Vowel-l sequences, on the other hand, did split apart; errors involved the l alone or the vowel alone, significantly more often than they involved the r or V alone from Vr sequences (Wilcoxon p <.0001). This supported the hypothesis that the r and l belonged to different constituents. The relative frequency of errors is commonly assumed to reflect constituency. See, for instance, Stemberger and Treiman’s (1986) arguments regarding different positions in the onset. Nevertheless, such sequences also acted as single units, and in this respect behaved like vowel-r sequences. In addition, when such sequences were involved as a unit they interacted with single vowels or vowel glide sequences, suggesting that at some level of analysis they belonged to the same constituents as vowels. Speech errors which involve interactions between two items, or the substitution of one item for another, are in general structure-preserving. They involve units of the same constituent type at some level of analysis; for instance, noun phrase with noun phrase, preposition with preposition, verb with verb, and at the sub-lexical level, syllabic constituent type a with syllabic constituent type a (Laubstein 1987).