The extent to which cellular metabolites are NMR observable is of fundamental importance in the interpretation of in vivo NMR studies. Analysis of ischemic rat liver shows that ATP resonances measured by 31P NMR decrease considerably faster than total tissue ATP measured in extracts. This discrepancy demonstrates that, in liver, ATP is not 100% observable. Furthermore, the data are consistent with the supposition that in situ mitochondrial ATP resonances are not normally observable by in vivo NMR techniques. The specificity of the NMR measurement for cytosolic ATP indicates that 31P NMR can be a valuable tool for the specific measurement of ATP in this compartment.