One-half century ago a classic experiment by G.D. Scott showed that pouring balls into a rigid container filled the volume to anupper limit of 64% of the container volume, well below the 74% volume fraction filled by spheres in a densest-possible packing. Many subsequent experiments have confirmed the “random closed packed” volume fraction limit. However, the physics of the random-closed-pack limit has remained a mystery. In an experiment on spheres in a box, we observed that with weak shearing (by slightly tilting two opposite walls), a volume fraction of 64.5% was reached and persisted for about 50000 shear cycles. Then the first growing nucleus appeared in the interior of the packing, indicating the onset of a first order phase transition, where isolated dense clusters with crystalline symmetry emerge in the interior of the less-dense amorphous bulk. As the shearing continued most nuclei with ten or fewer spheres dissolved, while larger ones grew. Previous experiments have seen crystallites appear on container walls but not in the bulk. Nucleation is of widespread interest, for instance in modeling glasses, which successfully resist nucleation. The distinctive qualitative features observed in our experiment should help guide the development of a theory of nucleation.