Abstract. Reduced-order modeling is an efficient approach for solving parameterized discrete partial differential equations when the solution is needed at many parameter values. An offline step approximates the solution space and an online step utilizes this approximation, the reduced basis, to solve a smaller reduced problem at significantly lower cost, producing an accurate estimate of the solution. For nonlinear problems, however, standard methods do not achieve the desired cost savings. Empirical interpolation methods represent a modification of this methodology used for cases of nonlinear operators or nonaffine parameter dependence. These methods identify points in the discretization necessary for representing the nonlinear component of the reduced model accurately, and they incur online computational costs that are independent of the spatial dimension N . We will show that empirical interpolation methods can be used to significantly reduce the costs of solving parameterized versions of the Navier-Stokes equations, and that iterative solution methods can be used in place of direct methods to further reduce the costs of solving the algebraic systems arising from reduced-order models.Key words. reduced basis, empirical interpolation, iterative methods, preconditioning 1. Introduction. Methods of reduced-order modeling are designed to obtain the numerical solution of parameterized partial differential equations (PDEs) efficiently. In settings where solutions of parameterized PDEs are required for many parameters, such as uncertainty quantification, design optimization, and sensitivity analysis, the cost of obtaining high-fidelity solutions at each parameter may be prohibitive. In this scenario, reduced-order models can often be used to keep computation costs low by projecting the model onto a space of smaller dimension with minimal loss of accuracy.We begin with a brief statement of the reduced basis method for constructing a reduced-order model. Consider an algebraic system of equations G(u) = 0 where u := u(ξ) is an unknown vector of dimension N , and ξ is vector of m input parameters. We are interested in the case where this system arises from the discretization of a PDE and N is large, as would be the case for a high-fidelity discretization. We will refer to this system as the full model. We would like to compute solutions for many parameters ξ. Reduced basis methods compute a (relatively) small number of full model solutions, u(ξ 1 ), . . . u(ξ k ), known as snapshots, and then for other parameters, ξ = ξ j , construct approximations of u(ξ) in the space spanned by {u(ξ j )}