In this paper, we present a new approach to handset identification using Support Vector Machines (SVMs) [1]. The inconsistency of audio characteristics among different handsets significantly degrades the performance of speaker recognition [2]. If a speaker recognizer can identify the handset a speaker is using, it can perform the recognition by selecting a model trained specifically on that handset. We present an SVM-based handset identifier that uses the Gaussian kernel and the one-vs-rest approach [1] to separate utterances on one kind of handset from those on the others. We analyze the performance of speaker-dependent and speaker-independent identifiers in classifying 4 different types of handset: carbon-button, electret, cordless, and headset. The test results show that SVMs yield greater than 90% accuracy, and that both speaker-dependent and speaker-independent approaches give comparable results on all test sets. Experiments also show the advantages of SVMs over the previous channel compensation (RASTA [3]) and handset identification (GMM/ML [2, 4]) algorithms.