SIM-Lite is an astrometric interferometer being designed for sub-microarcsecond astrometry, with a wide range of applications from searches for Earth-analogs to determining the distribution of dark matter. SIM-Lite measurements can be limited by random and systematic errors, as well as astrophysical noise. In this paper we focus on instrument systematic errors and report results from SIM-Lites interferometer testbed. We find that, for narrow-angle astrometry such as used for planet finding, the end-of-mission noise floor for SIM-Lite is below 0.035 uas.