The steady three-dimensional exterior flow of a viscoelastic non-Newtonian fluid is approximated by reducing the corresponding nonlinear elliptic-hyperbolic system to a bounded domain. On the truncation surface with a large radius R, nonlinear, local second-order artificial boundary conditions are constructed and a new concept of an artificial transport equation is introduced. Although the asymptotic structure of solutions at infinity is known, certain attributes cannot be found explicitly so that the artificial boundary conditions must be constructed with incomplete information on asymptotics. To show the existence of a solution to the approximation problem and to estimate the asymptotic precision, a general abstract scheme, adapted to the analysis of coupled systems of elliptic-hyperbolic type, is proposed. The error estimates, obtained in weighted Sobolev norms with arbitrarily large smoothness indices, prove an approximation of order O(R −2+ε ), with any ε>0. Our approach, in contrast to other papers on artificial boundary conditions, does not use the standard assumptions on compactly supported right-hand side f , leads, in particular, to pointwise estimates and provides error bounds with constants independent of both R and f . ARTIFICIAL BOUNDARY CONDITIONS FOR VISCOELASTIC FLOWS 939 variables. It must be stressed, however, that this is the only information available for the main asymptotic term.In principle, a similar idea could work for non-Newtonian flows. As for the asymptotic behaviour, the exterior problem for an Oldroyd-B fluid has already been studied in weighted spaces in [10].There it was shown that, although the Navier-Stokes equations are perturbed by the divergence of the extra-stress tensor, they maintain the (Navier-Stokes) asymptotic form. Hence, the ABCs for the Navier-Stokes part in the equations need only to be complemented with an additional term.The solution of the transport equation does not need to satisfy any boundary conditions and, therefore, at the first sight the ABCs stemming from the Navier-Stokes part seem to be enough for the approximation problem in R . This affirmation falls through, however, because the existence and uniqueness proofs for the transport equation (cf. [17, 18]) are crucially based on homogeneous Dirichlet data for the velocity field ‡ on the boundary * R which cannot of course happen on S R with our ABC. On the other hand, we know precisely the asymptotic decay properties of the viscoelastic solution at infinity by the results proven in [10]. The spontaneous idea of multiplying the transport term by a cut-off function leads to what we call an artificial transport equation (ATE) and, perhaps somewhat unexpectedly, we are able to prove that the error resulting from this multiplication is of the same order as the intrinsic error arising from the imposed ABC for the Navier-Stokes equations.To estimate the difference between the solutions in R and , we follow the approach proposed in [8] which both establishes the existence of a unique small solution to the proble...