Costly anti-predatory defences are used in ecological time and maintained in evolutionary time by natural selection favouring individuals that survive through their use. Autotomy of expendable body parts is a striking example of a defence having multiple substantial costs, including loss of ability to use the same defence, loss of energy, and decreased growth, reproductive success and survival following autotomy, plus the energetic cost of replacing the lost body part in species capable of regenerating them. Our study shows that autotomy in the lacertid lizard Podarcis lilfordi reduces sprint speed, indicating decreased capacity to escape as well as the loss of energy. Autotomy carries substantial cost, and thus should be avoided except as a last resort. Ease of autotomy and post-autotomic movements were studied in three populations of lacertid lizards. Two were islet populations of P. lilfordi from Aire (lowest predation pressure) and Colom (intermediate predation pressure) off Minorca. The third was a mainland population of Podarcis hispanica, a closely related species from the mainland of the Iberian Peninsula where predation pressure is higher than on the islets. As predicted, a suite of autotomic traits increases the effectiveness of autotomy as a defence as predation pressure increases. With increasing predation pressure, the frequency of voluntary autotomy increases, latency to autotomy decreases, pressure on the tail needed to induce autotomy decreases, vigour of post-autotomic tail movements increases, and distance moved by the shed tail increases. Additional changes that might be related to predation pressure, but could have other causes, are the presence of tail coloration contrasting with body coloration except under the lowest predation pressure (Aire) and longer tails in the mainland species P. hispanica. Correspondence between predation pressure and the suite of autotomic traits suggests that autotomy is an important defence that responds to natural selection. Comparative data are needed to establish the generality of relationships suggested in our study of only three populations.