Empirical mode decomposition (EMD), the core of Hilbert-Huang transformation, is reckoned on the algorithm of the extrama extending and generation of envelope curves. With the piecewise power function and an improved algorithm of extrama extending, a novel EMD algorithm was proposed. An error estimation of the piecewise power function interpolation was proposed to show the advantage of this algorithm. At last, a simulation on a stock model is presented to show that this novel EMD algorithm is more applicable to analyze the nonlinear and non-stationary signal than the previous EMD algorithm.