31P-NMR measurements of saturation transfer have been used to measure exchange between the gamma-phosphate of ATP and phosphocreatine and between the beta-phosphate of ATP and the beta-phosphate of ADP in the reaction catalysed by creatine kinase in vitro. The calculated exchange fluxes have been compared with measurements of 15N label exchange between creatine and phosphocreatine and 14C label exchange between ATP and ADP. At pH 8.0 the fluxes between phosphocreatine and the gamma-phosphate of ATP and between the beta-phosphates of ATP and ADP, measured by saturation transfer, were the same and equal, within experimental error, to the fluxes between creatine and phosphocreatine, measured by 15N label exchange, and between ADP and ATP, measured by 14C label exchange. At pH 7.0 the flux between phosphocreatine and the gamma-phosphate of ATP, measured by saturation transfer, was equal, within experimental error, to the flux between creatine and phosphocreatine, measured by 15N label exchange. However, at low ADP concentrations (less than 0.2 mM), the flux between ATP and ADP measured by saturation transfer was significantly less than that between phosphocreatine and ATP and, more importantly, less than the ADP-ATP exchange flux measured by 14C label exchange. The saturation transfer and isotope exchange measurements at pH 7.0 have shown that it is valid to equate saturation transfer measurements of exchange between phosphocreatine and ATP in vivo with the potential for net chemical flux through the reaction. The observed discrepancy at pH 7.0 between the 14C and saturation transfer measurements of ATP----ADP exchange can be explained if there is significant loss of saturation in an intermediate in the exchange reaction. Under these conditions analysis of the exchange according to two-site exchange model is invalid. In magnetisation transfer measurements of exchange in other enzyme catalysed reactions, the possible presence of a kinetically significant intermediate and therefore the validity of data analysis using a two-site exchange model should be considered.