Results are reported indicating the importance of spectral change in the identification of isolated vowels in Canadian English, including vowels usually considered monopthongal. Brief (30‐ms Hamming windowed) segments were extracted from the “steady‐state” (INITIAL) and “offglide” (FINAL) portions of 40 isolated vowels (ten categories × four speakers). Pairs of sections, separated by 10 ms, were presented to listeners in three conditions: (1) INITIAL‐FINAL (I‐F), (2) INITIAL‐INITIAL (I‐I), and (3) FINAL‐INITIAL (F‐I). Listeners' error rates were comparable for the I‐F condition (14%) and the original tokens (13%), but considerably higher for I‐I (32%) and F‐I (38%) conditions. The fundamental and the first two formant frequencies were measured in each of the “steady‐state” and “offglide” sections. These measures were used to predict listeners' identification using a model of the type described in Assmann, Nearey, and Hogan [J. Acoust. Soc. Am.71,975–989 (1982)]. Observed patterns of confusion were generally similar to model predictions.