Although crystallization is the most important method for the separation of enantiomers of chiral molecules in the chemical industry, the chiral recognition involved in this process is poorly understood at the molecular level. We report on the initial steps in the formation of layered racemate crystals from a racemic mixture, as observed by STM at submolecular resolution. Grown on a copper single-crystal surface, the chiral hydrocarbon heptahelicene formed chiral racemic lattice structures within the first layer. In the second layer, enantiomerically pure domains were observed, underneath which the first layer contained exclusively the other enantiomer. Hence, the system changed from a 2D racemate into a 3D racemate with enantiomerically pure layers after exceeding monolayer-saturation coverage. A chiral bias in form of a small enantiomeric excess suppressed the crystallization of one double-layer enantiomorph so that the pure minor enantiomer crystallized only in the second layer.