In a dynamic optimisation framework we show that the optimal tax evasion can be either a positive or a negative function of the tax rate according to the form of the fine to be paid when evasion is detected.