Sodium imaging is mainly performed with spin-density weighted sequences to quantify tissue sodium concentration. However, relaxation weighting can add additional specific information. We pursue appropriate sampling for low SNR and fast biexponential decay. The accuracy and precision of typical T2* measurements is determined for different sampling schemes by simulation and phantom measurements. We developed a dedicated sampling scheme for brain parenchyma employing numerical optimization. The results suggest that averaging is preferable to increase reliability compared to denser temporal sampling. In-vivo comparison confirmed the advantage of the optimized patter with increased separation of the decay components.