We propose an algorithm that can reconstruct 0.9mm3 isotropic T2 maps based on multiple 2D multi-echo spin-echo acquisitions that were highly undersampled. The data is reconstructed by combining a classical super-resolution approach with an iterative model-based reconstruction. Thereby, the reconstruction problem is split into multiple sub-problems to improve the convergence of the algorithm. Resulting T2 values within structures of the midbrain and the hippocampus from four healthy volunteers showed good reproducibility. This kind of high-resolution relaxometry may enable additional insight in pathologies of small brain structures and increased sensitivity to disease-induced changes.