The aim of this prospective, blinded, randomised controlled study was to compare novices' acquisition of the technical skills of ultrasound-guided regional anaesthesia using either a meat phantom model or fresh-frozen human cadavers. The primary outcome was the time taken to successfully perform an ultrasound-guided sciatic nerve block on a cadaver; secondary outcomes were the cumulative score of errors, and best image quality of the sciatic nerve achieved. After training, the median (IQR [range]) time taken to perform the block was 311(164–390 [68–600]) s in the meat model trained group and 210 (174–354 [85–600]) s in the fresh-frozen cadaver trained group (p = 0.24). Participants made a median (IQR [range]) of 18 (14–33 [8–55]) and 15 (12–22 [8–44]) errors in the two groups respectively (p = 0.39). The image quality score was also not different, with a median (IQR [range]) of 62.5 (59.4–65.6 [25.0–100.0])% vs 62.5 (62.5–75.0 [25.0–87.5])% respectively (p = 0.58). The training and deliberate feedback improved all participants' block performance, the median (IQR [range]) times being 310 (206–532 [110–600]) s before and 240 (174–354 [85–600]) s after training (p = 0.02). We conclude that novices taught ultrasound scanning and needle guidance skills using an inexpensive and easily constructed meat model perform similarly to those trained on a cadaveric model.