Many mathematical/computational models used in Modelling and Simulation require iterative solution procedures, as for example the mass and momentum balances solved in Computational Fluid Dynamics (CFD). Therefore, one of the contributions to the numerical error is the iterative error that may have several origins in CFD: segregated solution procedures; non-linearity of the equations; discretization techniques including deferred corrections and iterative techniques for the solution of the systems of algebraic equations. Iterative errors can only be estimated when a solution converged to machine accuracy is available, which may be extremely time consuming or even impossible in practical applications. Therefore, it is common practice to assess iterative errors using normalized residuals and/or differences between non-linear iterations which are not reliable estimates of the iterative error [1]. In this paper we illustrate again the shortcomings of using normalized residuals and/or differences between iterations to estimate iterative errors. Considering that mathematical models for CFD simulations are becoming mostly unsteady, the selected test case is the flow around a two-dimensional circular cylinder at Reynolds numbers of $Re=10^2$ and $Re=10^8$. The flow is laminar for the lowest Reynolds number and ensemble-averaged Reynolds equations using the Shear-Stress Transport (SST) k-w eddy-viscosity, two-equation turbulence model is selected for the highest Reynolds number. Implicit time integration is used in both cases and so the role of Courant number in the relation between iterative errors and normalized residuals and/or differences between consecutive iterations is illustrated for different levels of space and time refinement. Finally, the paper presents and tests a practical procedure for the estimation of iterative errors that does not require solutions converged to machine accuracy. The performance of the procedure is illustrated for the selected test case. Furthermore, we illustrate the interaction between iterative errors and the estimation of the time/space discretization errors using grid/time refinement studies.