A number of languages have independently developed a pattern of vowel-length neutralisation according to which a vowel after a vocoid must be long. It is proposed that this pattern arises from the inherent acoustic ambiguity of such sequences, which are realised with a diphthongal transition from one formant pattern to the next, with no clear boundary between the two. Neutralisation in vocoid sequences originates from listeners' difficulties in determining the duration of vowels in this context. Lengthening of the second vocoid arises when listeners attribute some of the transition duration to that segment. The phonetic bases of this account are supported by three experiments with Finnish speakers. A production study shows that speakers treat the transition as belonging in part to the realisation of the postvocoid vowel. Two perception studies show that increasing the duration of the transition increases the probability of such a vowel being identified as long.