Deepfake images have raised significant security concerns due to their malicious use. Deepfake detection algorithms based on deep learning have achieved promising performance. However, the algorithms face challenges in real-world deployments, especially in digital forensic applications. Therefore, it is essential to develop evaluation methods for deepfake detectors that characterize their reliability, security and efficiency. This chapter specifies practical benchmarks for evaluating deepfake detectors. An evaluation platform for deepfake image detectors is employed. The platform, comprising a sample generation engine, an interaction engine and a real-time analysis engine, significantly reduces the manual analysis efforts. A comprehensive, multidimensional quantitative evaluation system for deepfake detectors is described. The evaluation system has three primary metrics, ten secondary metrics and 21 tertiary metrics that provide holistic evaluations of detection, security and efficiency. Extensive experiments demonstrate the usability of the platform and the effectiveness of the evaluation system.