The past decade + of various NCEP Eta Model performance and verification tests is reviewed, including in particular the period following the recent extensions to 60 and then 84 h, with a view to indications if any as to what Eta dynamical core and/or setup features might be responsible for its relatively successful performance. Very extensive comparison tests against the fourth-order accurate NGM, and the "infinite" order accurate RSM in the early and mid- to later nineties are recalled, as well as the long-term parallel running of the "Early", 48 km/38 layer, and "Meso", 29 km/50 layer Eta. Inferences regarding the benefits from high Taylor-series accuracy, and from resolution vs domain size are implied. Since the operational Eta is driven by the lateral boundary condition from the previous run of the global (Avn) model, presently at the "on" times (00 and 12z) estimated to represent about an 8 h loss in accuracy, one might expect the skill of the Eta at later forecast times to fall behind that of the Avn. Three aspects of the Eta skill, relative to that of the Avn, are looked into for evidence of this taking place: precipitation scores, rms fits to raobs, and the accuracy in placing the centers of major storms at later forecast times. In none of these signs of the deterioration of the Eta skill compared to that of the Avn are found evident. Features are therefore felt necessary to be present in the Eta able to compensate for the inflow of the less accurate Avn boundary condition. Recalling the result of a recent eta/sigma experiment, and referred to and other tests of the high Taylor series accuracy approaches, it is pointed out that the eta coordinate, and the Eta's various Arakawa-style finite-volume-like features, are strong candidates in this sense. It is noted that the latter, the Arakawa-style features, mitigates the inconsistency that exists between the treatment of dynamics and "physics" in current NWP and climate models, of treating grid point values as point values of smooth functions in dynamics, as opposed to grid-box averages in physical parameterization schemes.