This project's goal is to design a SPECT insert for a clinical MRI system for simultaneous brain SPECT/MR imaging, with a high-sensitivity collimator and high-resolution detectors. We have compared eight collimator designs, four multi-pinhole and four multi-slit slit-slat configurations. The collimation was designed for a system with 2 rings of 25 5 x 5 cm detectors. We introduce the concept of 1/2-pinhole and 1/2-slit, which are transaxially shared between two adjacent detectors. Analytical geometric efficiency was calculated for an activity distribution corresponding to a human brain and a range of intrinsic detector resolutions R-i and target resolutions R-t at the centre of the FOV. Noise-free data were simulated with and without depth-of-interaction (DOI) information, 0.8 mm R-i and 10 mm R-t FWHM, and reconstructed for uniform, Defrise, Derenzo, and Zubal brain phantoms. Comparing the multi-pinhole and multi-slit slit-slat collimators, the former gives better reconstructed uniformity and transaxial resolution, while the latter gives better axial resolution. Although the 2 x 2-pin-hole and 2-slit designs give the highest sensitivities, they result in a sub-optimal utilisation of the detector FOV. The best options are therefore the 5+2 1/2-pinhole and the 1+2 1/2-slit systems, with sensitivities of 1.8 x 10(-4) and 3.2 x 10(-4), respectively. Noiseless brain phantom reconstructions with the multi-pinhole collimator are slightly superior as compared to slit-slat, in terms of symmetry and accuracy of the activity distribution, but the same is not true when noise is included. DOI information reduces artefacts and improves uniformity in geometric phantoms. Further evaluation is needed with prototype collimators.