Abstract: To deal with time-varying, nonlinear and low frequency characteristics of pulse signal, a method based on wavelet transform modulus maxima is presented. The three parameters of wavelet transform modulus maxima namely the Lipschitz exponent α, smooth factor σ and fractal dimension D(α) are introduced, and a filtering algorithm of wavelet transform modulus maxima curve is put forward. Experiments are performed on a three peaks pulse signal using Gaussian first-order derivative and second-order derivative wavelet transforms. Experimental results show that some characteristics that cannot be described in time domain can be extracted using this method, and the local feature points of pulse signal can be located precisely. It establishes the foundation of the quantitative description of the pulse characteristics and the recognition of pulse.