With respect to three kinds of familiar tremor

including essential tremor

parkinsonism disease tremor and physiological tremor

which are subjected to frequent clinical misdiagnosis

a new recognition approach for tremor based on bispectrum analysis and support vector machine is proposed in this paper.At first

the acceleration signals of hand tremor from voluntary subjects were recorded and were tested by Hinich method respectively

we found that the signals possess the properties of non-gaussian and non-linearity.Then the features of diagonal slice of bispectrum which adapts to process non-gaussian and non-linearity signals of hand tremor accelerating signals were extracted.Finally

multiclassification support vector machine algorithm of "one against one" and "one against rest" are adopted to carry out recognition of three types tremor.Cross-validation test results show that the mean correct rate of classification with "one against one" algorithm is better than that of "one against rest"

its correct rate of classification can be readod at 93.13%

and provides a new a assistant approach to classify tremor for clinical neurosurgeon.