Marine plastic pollution is a global problem with considerable ecological and economic consequences. Quantifying the amount of plastic in the ocean has been facilitated by surveys of accumulated plastic on beaches, but existing monitoring programmes assume the proportion of plastic detected during beach surveys is constant across time and space. Here we use a multi-observer experiment to assess what proportion of small plastic fragments is missed routinely by observers, and what factors influence the detection probability of different types of plastic. Detection probability across the various types of plastic ranged from 60 to 100%, and varied considerably by observer, observer experience, and biological material present on the beach that could be confused with plastic. Blue fragments had the highest detection probability, while white fragments had the lowest. We recommend long-term monitoring programmes adopt survey designs accounting for imperfect detection or at least assess the proportion of fragments missed by observers.