This paper introduces a multi-fidelity optimization method using co-kriging, an extension of kriging that incorporates correlated Gaussian processes. The method aims to improve the accuracy of surrogate models by combining data from multiple levels of analysis, each with varying computational costs and fidelity. The authors present an exchange algorithm to select points for sampling within each level of analysis and derive co-kriging equations, including a new variance estimator to account for computational noise. The methodology is demonstrated through a multi-fidelity wing optimization problem, where an aircraft wing design is optimized using both a fast empirical drag estimation code (Tadpole) and a slower linearized potential method (VSaero). The results show that the co-kriging method consistently outperforms a kriging-based approach, finding better optima with fewer evaluations and fewer failed simulations. The paper also discusses the use of regression constants in the co-kriging formulation to filter noise and the selection of sampling plans to ensure uniform coverage of the design space.This paper introduces a multi-fidelity optimization method using co-kriging, an extension of kriging that incorporates correlated Gaussian processes. The method aims to improve the accuracy of surrogate models by combining data from multiple levels of analysis, each with varying computational costs and fidelity. The authors present an exchange algorithm to select points for sampling within each level of analysis and derive co-kriging equations, including a new variance estimator to account for computational noise. The methodology is demonstrated through a multi-fidelity wing optimization problem, where an aircraft wing design is optimized using both a fast empirical drag estimation code (Tadpole) and a slower linearized potential method (VSaero). The results show that the co-kriging method consistently outperforms a kriging-based approach, finding better optima with fewer evaluations and fewer failed simulations. The paper also discusses the use of regression constants in the co-kriging formulation to filter noise and the selection of sampling plans to ensure uniform coverage of the design space.