The aim of this work is to establish objective methods to quantitatively analyze e-cigarettes’ sweetness. Sixty samples of e-liquid were collected and measured by electronic tongue. Based on correlation analysis of electronic tongue data and artificial sensory data, three sweetness evaluation models were established by partial least squares, artificial neural network and support vector machine. Comparison results indicated that the support vector machine model was the most reliable for predicting sweetness of unknown e-cigarette samples. The correlation coefficient of the model was 0.96 with average relative error of predicted results of 7.30% and root mean square error of predicted results of 0.61. It was concluded that the evaluation model based on the combination of electronic tongue and the support vector machine can achieve reliable prediction of unknown e-cigarettes’ sweetness.