The Kriging predictor is regularly used as a metamodel to reduce the cost of forward propagation of uncertainty. Such metamodels (or surrogates) are difficult to design for industrial-scale systems and can be expensive to develop with the required fidelity to physical models, such as finite elements. A compromise between fidelity and computational expense is therefore necessary and concerns are raised on the correctness of model updating results obtained using metamodels. In this paper the effect of code uncertainy on model updating using the Kriging predictor is studied. There are two aspects of code uncertainty: (1) arising from an inadequate number of design points and (2) the Gaussian process as part of the formulation of the Kriging predictor. The concern is that the neglect of code uncertainty could invalidate the updating parameters if they are to be used subsequently in a FE model. The effect of both forms of code uncertainty is studied in a simulated experiment based on a physical structure.