We develop a high dimensional nonparametric classification method named sparse additive machine (SAM), which can be viewed as a functional version of support vector machine (SVM) combined with sparse additive modeling. the SAM is related to multiple kernel learning (MKL), but is computationally more efficient and amenable to theoretical analysis. In terms of computation, we develop an efficient accelerated proximal gradient descent algorithm which is also scalable to large datasets with a provable O(1/k2 ) convergence rate, where k is the number of iterations. In terms of theory, we provide the oracle properties of the SAM under asymptotic frameworks. Empirical results on both synthetic and real data are reported to back up our theory.