Sampling with replacement involves drawing samples from a dataset, but without removing it from said dataset. This introduces the possibility of drawing the same sample twice or more, but is often handy as a way of generating several "different" sets from only one.