Before a cell divides, it copies all of its DNA, so each new cell gets a full set. Base pairing tells the cell what each new strand has to be.
An enzyme called helicase unzips the helix. It breaks the hydrogen bonds between the bases, and the covalent bonds along each strand stay whole. Each old strand then serves as a template, a pattern for a new strand. Wherever the old strand has A, the new strand gets T, and wherever it has G, the new strand gets C.
DNA polymerase
The enzyme that builds the new strand is DNA polymerase. It moves along the old strand, picks the building block whose base pairs with the next base, and joins it on. The building blocks arrive carrying three phosphate groups, like ATP. As each one joins, two of its phosphates break off, and that gives out the energy for the new link. Each new building block joins the -OH at the 3' end of the growing strand, so a new strand always grows from 5' to 3'.
DNA polymerase can only add to a strand that is already there, so each new strand starts from a short starting piece, a primer, made by another enzyme. In bacteria, DNA polymerase adds about 1,000 bases a second. It checks each new base and takes out most wrong ones, and with the cell's repair enzymes, about one mistake gets through in every billion bases copied.
One old strand in each copy
So each new helix should keep one old strand and gain one new one. In 1958 Matthew Meselson and Franklin Stahl, at the California Institute of Technology, showed that it does. They grew bacteria on food whose nitrogen was all nitrogen-15, a heavier kind of nitrogen atom with an extra neutron, until all their DNA was heavy. Then they moved the bacteria to food with ordinary nitrogen-14 and let them copy their DNA once, then twice.
Spun fast in a centrifuge in a dense salt solution, DNA settles at the height that matches its density. After one copy, all the DNA settled halfway between heavy and light: every helix had one heavy strand and one light one. After two copies, half settled halfway and half at light.