We test competing models that aim at explaining the nature of stars in the Milky Way that are well away (|z|$gtrsim$ 1kpc) from the midplane, the so-called thick disk: the stars may have gotten there through orbital migration, through satellite mergers and accretion, or through heating of pre-existing thin disk stars. Sales et al. (2009) proposed the eccentricity distribution of thick disk stars as a diagnostic to differentiate between these mechanisms. Drawing on SDSS DR7, we have assembled a sample of 34,223 G-dwarfs with 6-D phase-space information and metallicities, and have derived orbital eccentricities for them. Comparing the resulting eccentricity distributions, p(e|z), with the models, we find that: a) the observed p(e|z) is inconsistent with that predicted by orbital migration only, as there are more observed stars of high and of very low eccentricity; b) scenarios where the thick disk is made predominantly through abrupt heating of a pre-existing thin disk are also inconsistent, as they predict more high-eccentricity stars than observed; c) the observed p(e|z) fits well with a gas-rich merger scenario, where most thick disk stars were born from unsettled gas in situ.