Two experiments were carried out to assess the ability of lay listeners to identify familiar voices from shouted samples. The experiments were conducted after a murder case in which a witness claimed to recognise the voice of a masked gunman based on two shouted words: get him! Our experiments were designed to explore the extent to which a listener can identify a known voice from a shouted sample, and also whether a two word sample is sufficient for identification. Recordings were obtained from a group of nine females who formed a close social network, plus six foils. Two shouted utterances were extracted for listening tests: for Test 1, ‘get him!’, and for Test 2, ‘face down on the ground and hands behind your back now!’ Thirteen listeners from the same social network participated in the tests. Listeners correctly identified familiar speakers in 52% of cases in Test 1 and 81% in Test 2. There was considerable variation in results across listeners, and also with respect to individual voices. The results suggest that recognition of shouted voices is far from perfect, even in closed tests carried out among a close network. The variability in performance emphasises the need to subject a witness’s ability to identify a voice in a forensic case to formal testing wherever this is feasible.