Magnetic resonance spectroscopy (MRS) can give information about cellular metabolism in vivo which is difficult to obtain in other ways. In skeletal muscle, non-invasive (31) P MRS measurements of the post-exercise recovery kinetics of pH, [PCr], [Pi] and [ADP] contain valuable information about muscle mitochondrial function and cellular pH homeostasis in vivo, but quantitative interpretation depends on understanding the underlying physiology. Here, by giving examples of the analysis of (31) P MRS recovery data, by some simple computational simulation, and by extensively comparing data from published studies using both (31) P MRS and invasive direct measurements of muscle O2 consumption in a common analytical framework, we consider what can be learnt quantitatively about mitochondrial metabolism in skeletal muscle using MRS-based methodology. We explore some technical and conceptual limitations of current methods, and point out some aspects of the physiology which are still incompletely understood.