Catastrophic cancellation occurs in numerical computing when subtracting two nearly equal numbers results in significant loss of precision. This phenomenon is particularly problematic in floating-point arithmetic, where it can lead to large relative errors in calculations that are otherwise expected to be accurate.