This paper presents a multi-fidelity optimization approach using co-kriging, a statistical method that combines data from multiple levels of analysis to improve the accuracy and efficiency of surrogate models. The method uses an exchange algorithm to select points for sampling at each level of analysis, and a new variance estimator to account for varying degrees of computational noise. The approach is demonstrated using a multi-fidelity wing optimization problem, where a high-fidelity simulation is combined with a low-fidelity empirical model to improve the accuracy of the optimization. The paper discusses the use of co-kriging, an extension of kriging, to model the relationship between high- and low-fidelity data. The method involves estimating hyper-parameters and scaling factors to account for the differences between the data sets. The co-kriging model is used to predict the high-fidelity function values, and the error estimates are adjusted to reflect the uncertainty in the predictions. The paper also addresses the issue of sampling plans, where the selection of points for high- and low-fidelity simulations is optimized to ensure uniform coverage of the design space. The method is applied to a transonic aircraft wing optimization problem, where a low-fidelity empirical drag estimation code is combined with a high-fidelity physics-based simulation to minimize drag. The results show that the co-kriging method outperforms traditional kriging in terms of accuracy and efficiency, finding better optima with fewer simulations and fewer failed runs. The paper concludes that the use of correlated surrogates can significantly enhance the accuracy of surrogate models, and that the co-kriging method provides a robust and efficient approach for multi-fidelity optimization. The method is applicable to a wide range of engineering design problems where multiple levels of analysis are available.This paper presents a multi-fidelity optimization approach using co-kriging, a statistical method that combines data from multiple levels of analysis to improve the accuracy and efficiency of surrogate models. The method uses an exchange algorithm to select points for sampling at each level of analysis, and a new variance estimator to account for varying degrees of computational noise. The approach is demonstrated using a multi-fidelity wing optimization problem, where a high-fidelity simulation is combined with a low-fidelity empirical model to improve the accuracy of the optimization. The paper discusses the use of co-kriging, an extension of kriging, to model the relationship between high- and low-fidelity data. The method involves estimating hyper-parameters and scaling factors to account for the differences between the data sets. The co-kriging model is used to predict the high-fidelity function values, and the error estimates are adjusted to reflect the uncertainty in the predictions. The paper also addresses the issue of sampling plans, where the selection of points for high- and low-fidelity simulations is optimized to ensure uniform coverage of the design space. The method is applied to a transonic aircraft wing optimization problem, where a low-fidelity empirical drag estimation code is combined with a high-fidelity physics-based simulation to minimize drag. The results show that the co-kriging method outperforms traditional kriging in terms of accuracy and efficiency, finding better optima with fewer simulations and fewer failed runs. The paper concludes that the use of correlated surrogates can significantly enhance the accuracy of surrogate models, and that the co-kriging method provides a robust and efficient approach for multi-fidelity optimization. The method is applicable to a wide range of engineering design problems where multiple levels of analysis are available.