Shape memory alloys (sma) are more and more integrated in engineering applications. These materials with their shape memory effect permit to simplify mechanisms and to reduce the size of actuators. sma parts can easily be activated by Joule effect but their modelling and consequently their control remains difficult, it is principally due to their hysteretic thermomechanical behaviour. Most of successful control strategy applied to sma actuator are not often suitable for industrial applications: they are particularly heavy and use the Preisach model or neural networks to model the hysteretic behaviour of these material; this kind of models are difficult to identify and to use in real time. That is why this paper deals with an application of the new framework of model-free control (mfc) to a sma spring based actuator. This control strategy is based on new results on fast derivatives estimation of noisy sig- nals, its main advantages are: its simplicity and its robustness. Experimental results and comparisons with pi control are exposed that demonstrate the efficiency of this new control strategy. Key words: Nonlinear control, Model-free control, Shape memory alloy, Derivative estimation, Nonphysical modelling.