Molecular gut-content analysis enables detection of arthropod predation with minimal disruption of ecosystem processes. Mass-collection methods, such as sweep-netting, vacuum sampling and foliage beating, could lead to regurgitation or rupturing of predators along with uneaten prey, thereby contaminating specimens and compromising resultant gut-content data. Proponents of this ‘cross-contamination hypothesis’ advocate hand-collection as the best way to avoid cross-contamination. However, hand-collection is inefficient when large samples are needed, as with most ecological research. We tested the cross-contamination hypothesis by setting out onto potato plants immature Coleomegilla maculata and Podisus maculiventris that had been fed larvae of either Leptinotarsa decemlineata or Leptinotarsa juncta, or unfed individuals of these predator species along with L. decemlineata larvae. The animals were then immediately re-collected, either by knocking them vigorously off the plants onto a beat cloth and capturing them en masse with an aspirator (‘rough’ treatment) or by hand-searching and collection with a brush (‘best practice’). Collected predators were transferred in the field to individual vials of chilled ethanol and subsequently assayed by PCR for fragments of cytochrome oxidase I of L. decemlineata and L. juncta. Ten to 39 per cent of re-collected fed predators tested positive by PCR for DNA of both Leptinotarsa species, and 14–38% of re-collected unfed predators contained L. decemlineata DNA. Overall levels of cross-contamination in the rough (31%) and best-practice (11%) samples were statistically different and supported the cross-contamination hypothesis. A pilot study on eliminating external DNA contamination with bleach prior to DNA extraction and amplification gave promising results.