The low signal-to-noise ratio (SNR) is the single main factor limiting the spatial resolution that can be attained in nuclear magnetic resonance spectroscopy (MRS) in animals or man. This work first reviews the fundamental relationship between metabolite concentration, spatial resolution and SNR, which together define the frontier of sensitivity for MRS. It then presents SLOOP (Spectral Localization with Optimal Pointspread-function) as a recipe to identify an optimal set of experimental parameters that will provide the most accurate results, in situations where geometrical a priori information about the sample is available. SLOOP, which provides robust, quantifiable spectroscopic information, has mainly been applied to study cardiac energy metabolism in healthy volunteers and in patients, and these results are briefly summarized. The quality of localized spectroscopy may be further improved through the use of nonlinear pulsed magnetic field gradients. This technology, which has been named SLOOP^N , is outlined and discussed. The text provides advice on the practical implementation of these methods with a focus on the high experimental quality required for localized MRS: even the most sophisticated postprocessing algorithm cannot fully compensate shortcomings in data acquisition.