Formant frequencies of the semivowels /j/ and /w/ in Amharic, Yoruba and Zuni were measured in three vowel environments. Cross-language differences were found between what are described as the same semivowels, i.e. different languages have different acoustic targets for /j/ and /w/. These cross-language differences in semivowels correlate with cross-language differences in the respective cognate vowels /i/ and /u/. Nonetheless, the semivowels differ in systematic ways from the vowels in directions that make them more 'consonantal'. These languages also differ in their patterns of coarticulation between semivowels and adjacent vowels. This shows, inter alia, that palatal segments differ from language to language in their degree of resistance to coarticulation. Because of these language-specific coarticulatory patterns, cross-language differences in acoustic targets can only be established after careful consideration of the effect of context.