We perform direct large molecular dynamics simulations of homogeneous SPC/E water nucleation, using up to˜4 ṡ106 molecules. Our large system sizes allow us to measure extremely low and accurate nucleation rates, down to˜1019 cm-3 s-1, helping close the gap between experimentally measured rates˜1017 cm-3 s-1. We are also able to precisely measure size distributions, sticking efficiencies, cluster temperatures, and cluster internal densities. We introduce a new functional form to implement the Yasuoka-Matsumoto nucleation rate measurement technique (threshold method). Comparison to nucleation models shows that classical nucleation theory over-estimates nucleation rates by a few orders of magnitude. The semi-phenomenological nucleation model does better, under-predicting rates by at worst a factor of 24. Unlike what has been observed in Lennard-Jones simulations, post-critical clusters have temperatures consistent with the run average temperature. Also, we observe that post-critical clusters have densities very slightly higher,˜5%, than bulk liquid. We re-calibrate a Hale-type J vs. S scaling relation using both experimental and simulation data, finding remarkable consistency in over 30 orders of magnitude in the nucleation rate range and 180 K in the temperature range. Homogeneous SPC/E water nucleation in large molecular dynamics simulations We perform direct large molecular dynamics simulations of homogeneous SPC/E water nucleation, using up to ∼4 · 10 6 molecules. Our large system sizes allow us to measure extremely low and accurate nucleation rates, down to ∼10 19 cm −3 s −1 , helping close the gap between experimentally measured rates ∼10 17 cm −3 s −1 . We are also able to precisely measure size distributions, sticking efficiencies, cluster temperatures, and cluster internal densities. We introduce a new functional form to implement the Yasuoka-Matsumoto nucleation rate measurement technique (threshold method). Comparison to nucleation models shows that classical nucleation theory over-estimates nucleation rates by a few orders of magnitude. The semi-phenomenological nucleation model does better, under-predicting rates by at worst a factor of 24. Unlike what has been observed in Lennard-Jones simulations, post-critical clusters have temperatures consistent with the run average temperature. Also, we observe that post-critical clusters have densities very slightly higher, ∼5%, than bulk liquid. We re-calibrate a Hale-type J vs. S scaling relation using both experimental and simulation data, finding remarkable consistency in over 30 orders of magnitude in the nucleation rate range and 180 K in the temperature range. C 2015 AIP Publishing LLC. [http://dx