Specific radar emitter identification, a.k.a. radar emitter individual identification, plays a momentous role in modern electronic countermeasures. This paper concentrates on the multiple kernel learning (MKL)-based multi-feature fusion methodology, and a multi-kernel discriminant analysis (MKDA) framework is thus proposed for radar individual identification. Based on the Doppler shift slices of ambiguity function (AF) representation, multiple kernels corresponding to the AF slices are constructed firstly. Then MKDA algorithms are utilized to extract the nonlinear discriminative features, followed by the simplest nearest neighbor classifier. Considering the efficiency of real-time systems, we devise two two-stage MKDA algorithms based on the criteria of kernel discriminant ratio (KDR) and kernel between-class scatter degree (KBCSD), respectively. We use two real radar datasets to evaluate the proposed framework. The KDR-MKDA and KBCSD-MKDA algorithms perform best in terms of both time efficiency and identification accuracy.