One of the important implications of a no-net-rotation (NNR) model of the Earth's horizontal surface velocity is that it is used as a constraint in the definition of the International Terrestrial Reference Frame (ITRF). We present here our most recent estimates of an NNR model (GSRM-NNR-2) that uses over 5700 geodetic station velocities. We test the sensitivity of our model by varying data input and model assumptions. The difference between our various models is maximally 0.6 mm yr-1, but can be larger if we use only a geographically limited core velocity data set, or if a significant bias in frame origin translation rate exists. Although plate boundary zones should not be considered as being part of rigid plates (Kreemer and Holt, 2001), the details of the velocity field within the plate boundaries have an insignificant effect on the NNR frame. The difference between the various NNR models that we test is small compared to the difference between our NNR models and ITRF2000, which can be up to 3.1 mm yr-1. We conclude that the ITRF2000 does not satisfy the NNR condition when considering an NNR model that is based on the actual present-day velocity of the entire Earth's surface, instead of NNR-NUVEL-1A.