We investigate bias in ground motions predicted for Central and Eastern North America (CENA) using ground motion models (GMMs) combined with site amplification models developed in the NGA-East project. Bias is anticipated because of de-coupled procedures used in the development of the GMMs and site amplification models. The NGA-East GMMs were mainly calibrated by adjusting CENA data to a reference site condition using a site amplification model appropriate for active tectonic regions. Hence, these GMMs are likely biased relative to the CENA reference site condition (3000 m/sec shear wave velocity). Moreover, the NGA-East site amplification model recommended for hazard applications contains a simulation-based term for amplification between the reference condition and time-averaged shear wave velocity VS30=760 m/sec, which is uncertain and has not been calibrated against data from sites with that reference condition. Using the NGA-East dataset, we apply mixed-effects residual analysis and identify that period-dependent bias in 5% damped response spectral acceleration is present across a wide frequency range, but is strongest (i.e., overestimating by a factor of 2) at short oscillator periods <0.2 sec. Ongoing work to remedy this bias consists of expanding the NGA-East dataset with more recent recordings and enhanced metadata, particularly regarding site conditions.