The basic concepts and methods on performance and effectiveness evaluations at the feature-level fusion model of multi-biometric encryption are concentrated on in this paper. From the cryptographic theory point of view, firstly, the formal definitions related to multi-biometric cryptosystems are formulated. Under some extreme conditions, the security and privacy of multi-biometric cryptosystems at the feature level are analyzed and rigorously proved. Finally, a close relationship between security and privacy and the fundamental tradeoff between the accuracy and security are studied.