This paper studies the deterrent effect of a medium increase in the severity of sentences and documents how the understanding of the newly enacted law evolves in the criminal population. The empirical strategy exploits mandatory sentencing act against repeat offenders passed by the French parliament in August 2007. The law specifically targets offenders who did the same act twice. Nevertheless, it was mainly presented as a law against recidivists without further explanation. I use a unique dataset including individual level data reporting the outcome of all trials ended with a conviction in France. The identification strategy is based on timing of events framework and competing risk analysis. It exploits the fact that offenders sentenced before the law was passed differ by the type of their new offense and the time between the first conviction and the date of application of the law. The analysis shows a significant short term deterrent effect of the law which disappear after some months. It also shows that the short term effect also concerns crimes that were presented as targeted but were not effectively affected. After four months, new offenses of different type are not deterred while new offenses of similar type still are.