The analysis and presentation of the results for the vowel features measured for the 60 speakers have been published in Wang & Van Heuven (2006, 2018) and also as part of a more comprehensive study involving additional native languages (Hungarian, Danish, Swedish, Norwegian) in Van Heuven (2016) and Van Heuven & Gooskens (2016). In order to determine the speaker’s L1, we now entered 30 vowel properties as predictors in a Linear Discriminant Analysis (LDA, Klecka 1980). Predictors were the durations of the ten monophthongs (after z-normalization within speakers) and the lowest two resonances of the vocal tract, i.e. F1 and F2. Formants were intrinsically normalized for the individual speaker’s vocal tract length (i.e., by dividing both F1 and F2 by F3 – see introduction). Running the LDA in stepwise mode, an optimal model was selected which yielded 80% correct determination of the speaker’s L1 using nine predictors: F2(ʊ), F1(ε), F1(ɪ), F1(e), D(ʊ), F1(ɒ), D(ɒ), F2(ε), and F1(o), in descending order of importance (see Table 1). Clearly, the spectral properties outweigh the duration cues, even though the latter were found to significantly increase the performance of the model. Spectral cues of lax (short) vowels contribute more to the correct prediction of the speaker’s L1 than those of tense (long) vowels. Table 2 presents the confusion matrix of the speakers’ predicted L1 (in the columns) against the actual L1 (in the rows), with the correct decisions on the main diagonal.