Crime has to be punished, but does punishment reduce crime? We conduct a neutrally framed laboratory experiment to test the deterrence hypothesis, namely that crime is weakly decreasing in deterrent incentives, i.e. severity and probability of punishment. In our experiment, subjects can steal from another participant's payoff. Deterrent incentives vary across and within sessions. The across subject analysis clearly rejects the deterrence hypothesis: except for very high levels of incentives, subjects steal more the stronger the incentives. We observe two types of subjects: selfish subjects who act according to the deterrence hypothesis and fair-minded subjects for whom deterrent incentives backfire.