Fully 3D Monte-Carlo (F3DMC) reconstruction consists in calculating a fully 3D object-specific system matrix using Monte-Carlo simulations and inverting it using an iterative approach. To reduce the large amount of disk space required by this approach, we derived a targeted F3DMC approach (TF3DMC) in which the volume to be reconstructed is irregularly sampled, so that pre-identified functional regions of interest are reconstructed using fine sampling while regions with non-specific activity or without any particular interest are coarsely sampled. This method was assessed using simulated and real SPECT data of a phantom filled with Tc99m. The GATE Monte-Carlo simulator was considered to simulate the phantom data and to calculate the system matrices needed for the reconstruction of the simulated and of the real SPECT data. Activity ratios measured in TF3DMC images were compared with those measured on F3DMC and OSEM images corrected for scatter, attenuation and detector response function. TF3DLMC yielded errors less than 10% in activity ratio estimates in hot regions, while errors with quantitative OSEM were between -21% and -3%. The space needed to store the system matrix was divided by a factor from 3.5 to 9.4 compared to F3DMC, for similar or even better accuracy in activity ratio estimates. These results suggest that TF3DMC can be made practical and outperforms F3DMC and OSEM in terms of quantitative accuracy.