A new strategy to yield information from the maximum number of voxels, each at the optimum signal-to-noise ratio (SNR) per unit time, in MR spectroscopic imaging (MRSI) is introduced. In the past, maximum acquisition duty-cycle was obtained by multiplexing in time several single slices each repetition time (TR), while optimal SNR was achieved by encoding the entire volume of interest (VOI) each TR. We show that optimal SNR and acquisition efficiency can both be achieved simultaneously by multiplexing in space and time several slabs of several slices, each. Since coverage of common VOIs in 3D proton MRSI in the human brain typically requires eight or more slices, at 3 T or higher magnetic fields, two or more slabs can fit into the optimum TR (ϳ1.6 s). Since metabolite concentrations in the brain are lower than tissue water by orders of magnitude, magnetic resonance spectroscopy (MRS) has been beset by inferior signal-to-noise ratio (SNR) per unit time compared to MRI. To partially compensate for this, MRS voxels are typically more than ϫ10 3 larger (cm 3 vs. mm 3 ), and acquisition times are ϫ10 2 -10 3 longer (many minutes vs. a few seconds) than MRI (1). It is not surprising, therefore, that substantial technical and methodological efforts have been made to improve the SNR and reduce the acquisition time of MRS, especially its most prevalent proton ( 1 H) variants. These efforts have progressed along two main paths. The first method involved multiplexing in space, i.e., increasing the spatial coverage from a single-to a simultaneous multivoxel acquisition (2,3). This technique (see Fig. 1a), which is known as MRS imaging (MRSI) and was first described for 1 H by Luyten et al. (4), yields much more spatial information at equal SNR (5), as determined by the voxel size and acquisition time (6). To reduce SNR loss from incomplete longitudinal relaxation, T 1 , the TR is extended beyond the actual acquisition-cycle, T C , the time needed to prepare and acquire one transient, as determined primarily by the desired spectral resolution. Since the system "recovers" for (TR-T C ), its duty cycle (DC) is an inherently suboptimal Ͻ100%. The second approach, introduced for 1 H-MRSI by Duyn and Moonen (7), was to maximize the amount of information per unit time by increasing the DC. This was achieved by multiplexing in time, acquiring several T C 's every TR, each from a different slice, as shown in Fig. 1b (8). The problem with this approach is that in order to satisfy spatial coverage and spectral resolution requirements, TR must be longer than the optimum, TR opt , and consequently the SNR is suboptimal.In this paper we extend the approach of 3D-multislab MRI (9) to 1 H-MRSI by multiplexing in space and time, as shown in Fig. 1c, and derive expressions for its gains in the DC and SNR over the current state of the art. We demonstrate that when more than one T C fits into TR opt , slab interleaving is more efficient than either the multislice approach (7) or whole volume of interest (VOI) per TR opt coverage (2,3,10 -15). W...