Besides the parenchymal diffusion and microvascular pseudo-diffusion, a third diffusion component was previously found in cerebral intravoxel incoherent motion (IVIM), representing interstitial fluid. However, estimating this intermediate IVIM component can be challenging, since spectral decomposition techniques have strong dependence on the number of samples (i.e. acquired b-values) and SNR. Therefore, it is important to know which b-values are essential to be acquired. In this study, an optimal b-value sampling for estimating the intermediate component is derived using a genetic algorithm. The optimal sampling (0,20,100,270,280,370,540,650,660,710,720,790,980,990,1000 s/mm2) was shown to outperform linear and logarithmic samplings, for both simulated and in vivo data.