A correction used in statistics when approximating from a discrete to a continuous distribution.
For example, when approximating from a binomial distribution or a Poisson distribution to a normal distribution, it is necessary to make a continuity correction (e.g. if X' is a normal approximation of the binomial variable X, then P(X<13) is equivalent to P(X'<12.5) assuming that the approximation is appropriate).
A continuity correction is needed because the area of a line is zero, so the correction is needed so that the calculated approximate probability is as close to the actual probability as possible.