In order to predict world crude oil price, an empirical mode decomposition (EMD) based support vector machine (SVM) ensemble learning paradigm is proposed. The original crude oil spot price series are first decomposed into a finite number of independent intrinsic mode functions (IMFs), with different frequencies. Then the IMFs are composed into three sub-series, namely fluctuating process, big events and a trend, based on fine-to-coarse reconstruction rule. Then different SVM models are used to model and forecast the three sub-series respectively. Finally, the forecasts of the three sub-series are combined with another SVM model to formulate an ensemble forecast for the original crude oil price series. To validate the proposed ensemble learning paradigm, two main crude oil price series, West Texas Intermediate (WTI) crude oil spot price and Brent crude oil spot price are used. The empirical results demonstrate effectiveness and attractiveness of the proposed EMDbased SVM ensemble learning paradigm compared with single SVMs and artificial neural networks.