We present the first microlensing candidate for a free-floating exoplanet-exomoon system, MOA-2011-BLG-262, with a primary lens mass of M_host ~ 4 Jupiter masses hosting a sub-Earth mass moon. The data are well fit by this exomoon model, but an alternate star+planet model fits the data almost as well. Nevertheless, these results indicate the potential of microlensing to detect exomoons, albeit ones that are different from the giant planet moons in our solar system. The argument for an exomoon hinges on the system being relatively close to the Sun. The data constrain the product M pi_rel, where M is the lens system mass and pi_rel is the lens-source relative parallax. If the lens system is nearby (large pi_rel), then M is small (a few Jupiter masses) and the companion is a sub-Earth-mass exomoon. The best-fit solution has a large lens-source relative proper motion, mu_rel = 19.6 +- 1.6 mas/yr, which would rule out a distant lens system unless the source star has an unusually high proper motion. However, data from the OGLE collaboration nearly rule out a high source proper motion, so the exoplanet+exomoon model is the favored interpretation for the best fit model. However, the alternate solution has a lower proper motion, which is compatible with a distant (so stellar) host. A Bayesian analysis does not favor the exoplanet+exomoon interpretation, so Occam's razor favors a lens system in the bulge with host and companion masses of M_host = 0.12 (+0.19 -0.06) M_solar and m_comp = 18 (+28 -100 M_earth, at a projected separation of a_perp ~ 0.84 AU. The existence of this degeneracy is an unlucky accident, so current microlensing experiments are in principle sensitive to exomoons. In some circumstances, it will be possible to definitively establish the low mass of such lens systems through the microlensing parallax effect. Future experiments will be sensitive to less extreme exomoons.